WEBVTT

1
00:00:00.151 --> 00:00:08.440
I think generative AI is at its heart con, and seeing these ultra-rich, ultra-powerful people lie through their f***ing teeth, turns my stomach.

2
00:00:08.520 --> 00:00:09.961
The word con is a strong word.

3
00:00:10.021 --> 00:00:16.428
Well, what do you call something where from the very beginning they've sold it in the terms of magic, but it's just a half-arsery machine?

4
00:00:16.488 --> 00:00:18.090
They are misleading the entire world,

5
00:00:18.290 --> 00:00:20.753
You are the first person that I've spoken to that has that opinion.

6
00:00:20.833 --> 00:00:25.138
Well, the fact that this is happening is insane, and the fact it's not a scandal is insane.

7
00:00:25.279 --> 00:00:30.145
And I've been in the tech industry for 16 years now, and I love technology, and I'm enthusiastic about it.

8
00:00:30.165 --> 00:00:32.447
But I don't like being misled.

9
00:00:32.467 --> 00:00:36.552
And this is the largest non-consensual push of technology in history.

10
00:00:36.693 --> 00:00:37.694
So we're going to play a game, Ed.

11
00:00:37.914 --> 00:00:42.098
I have the things that you consider to be myths about the AI industry.

12
00:00:42.219 --> 00:00:42.679
Let's play it.

13
00:00:43.180 --> 00:00:45.862
The AI industry is creating enormous economic growth.

14
00:00:46.083 --> 00:00:46.663
No, it's not.

15
00:00:46.703 --> 00:00:48.886
All of these companies run at a horrifying loss.

16
00:00:49.146 --> 00:00:51.628
OpenAI lost $20.9 billion last year.

17
00:00:51.829 --> 00:00:55.552
None of these people can just say, yeah, we're on the path to making this profitable because they can't.

18
00:00:55.673 --> 00:00:55.933
Next one.

19
00:00:56.273 --> 00:00:58.274
AI will replace all human jobs.

20
00:00:58.634 --> 00:01:01.755
That just isn't happening, and there's no economic data to support it.

21
00:01:01.915 --> 00:01:05.536
Next, the United States need to spend trillions to beat China in the AI race.

22
00:01:05.776 --> 00:01:09.037
What's the race to do, for us to constantly piss our pants worrying about China?

23
00:01:09.137 --> 00:01:11.818
But people keep saying, what if these models fall into the wrong hands?

24
00:01:11.898 --> 00:01:13.398
They're already in the wrong hands.

25
00:01:13.618 --> 00:01:16.119
Mark Zuckerberg, Sam Altman, Dario Amadei.

26
00:01:16.259 --> 00:01:20.721
Mark Zuckerberg says, we'll continue to invest aggressively in infrastructure to meet the demand.

27
00:01:20.821 --> 00:01:22.242
God matters a monstrosity.

28
00:01:22.542 --> 00:01:24.643
Makes me think of Shrek with Law Farquaad.

29
00:01:24.803 --> 00:01:27.424
Some of you may die, but that's a risk I'm willing to accept.

30
00:01:27.544 --> 00:01:33.526
If only these people gave a fuck about poverty or actual problems in the world versus are we buying enough GPUs.

31
00:01:33.866 --> 00:01:36.588
If this continues, what does the future look like?

32
00:01:36.608 --> 00:01:36.708
Fuck.

33
00:01:41.695 --> 00:01:43.976
Guys, I've got a favour to ask before this episode begins.

34
00:01:44.276 --> 00:01:50.337
The algorithm, if you follow a show, will deliver you the best episodes from that show very prominently in your feed.

35
00:01:50.637 --> 00:01:55.838
So when we have our best episodes on this show, the most shared episodes, the most rated episodes, I would love you to know.

36
00:01:55.998 --> 00:01:58.819
And the simple way for you to know that is to hit that follow button.

37
00:01:58.999 --> 00:02:03.280
But also it's the simple, easy, free thing that you can do to help us make the show better.

38
00:02:03.580 --> 00:02:08.721
And I would be hugely grateful if you could take a minute on the app you're listening to this on right now and hit that follow button.

39
00:02:16.027 --> 00:02:16.927
Ed Zitron.

40
00:02:18.109 --> 00:02:21.752
There are a number of things that you believe that a lot of other people don't believe.

41
00:02:22.073 --> 00:02:22.253
Right.

42
00:02:22.273 --> 00:02:30.121
You have, I think, a couple of controversial opinions and opinions that are in contrast to the other guests that I've sat here with.

43
00:02:30.862 --> 00:02:32.503
What exactly are those opinions, Ed?

44
00:02:32.523 --> 00:02:32.744
Ed Zitron.

45
00:02:33.471 --> 00:02:36.533
I think generative AI is at its heart con.

46
00:02:36.913 --> 00:02:40.015
I don't think it is sold as honest software.

47
00:02:40.115 --> 00:02:57.187
I think that they overstate both what it can do, what it will do, and the underlying financials to the point that they are misleading the entire world and they're actively exploiting the weaknesses in journalism, in our economies, and indeed within the responsible parties with sell-side analysts, governments, and all over the shop.

48
00:02:57.367 --> 00:02:58.788
The word con is a strong word.

49
00:02:59.475 --> 00:03:09.180
Yeah, I mean, what do you call something where from the very beginning they've sold it in the terms of magic as this thing that will replace all jobs, that will cure cancer, as all of these things.

50
00:03:09.200 --> 00:03:16.343
And when you look at it, it's boring cloud software that's extremely expensive and unprofitable and also unreliable at its core.

51
00:03:17.183 --> 00:03:24.168
People will be asking, where are you drawing from in terms of your references, your personal experiences, where will you educate, what you study, what you write about, what you do, Ed?

52
00:03:24.528 --> 00:03:27.830
So that's the funny thing is people say, he's not got a finance experience.

53
00:03:27.870 --> 00:03:28.630
He's not going to take.

54
00:03:28.811 --> 00:03:33.774
I've been in the tech industry 15, 16 years now in PR, but still had practical experience.

55
00:03:33.954 --> 00:03:36.615
And I love technology and I'm enthusiastic about it.

56
00:03:37.156 --> 00:03:38.277
And this thing just comes along.

57
00:03:38.417 --> 00:03:40.538
Everyone is telling me it's the best thing since sliced bread.

58
00:03:40.698 --> 00:03:42.159
It can't even do the basics.

59
00:03:42.199 --> 00:03:43.300
It can't even do search.

60
00:03:43.340 --> 00:03:46.102
Well, whenever you ask an AI person, well, what's your setup?

61
00:03:46.483 --> 00:03:50.906
They describe this Pee Wee's Playhouse thing of like, well, you've got a harness here and you've got to use the right prompt.

62
00:03:50.926 --> 00:03:52.488
Well, you don't want to use that prompt.

63
00:03:52.848 --> 00:03:58.352
You want to use this prompt here with this model, but don't use this model for the beginning, but at the end, you're going to want to use this model.

64
00:03:58.372 --> 00:03:59.093
But...

65
00:03:59.273 --> 00:04:01.756
And this is meant to be artificial intelligence.

66
00:04:01.916 --> 00:04:03.137
It's meant to be smart.

67
00:04:03.398 --> 00:04:04.719
It's meant to be autonomous.

68
00:04:04.799 --> 00:04:06.501
It's meant to be something that you set and forget.

69
00:04:06.761 --> 00:04:09.865
We have the sort of six leading AI companies on the table here.

70
00:04:10.105 --> 00:04:13.088
Anthropic, Amazon, NVIDIA, Microsoft, OpenAI, Google.

71
00:04:13.789 --> 00:04:16.191
You're saying that their fundamental business model is...

72
00:04:17.612 --> 00:04:19.873
Well, their revenues are not really coming from AI.

73
00:04:20.213 --> 00:04:24.274
Up until fairly recently, none of their revenues were coming from AI, like dribbles a bit.

74
00:04:24.674 --> 00:04:37.238
Right now, 70% of all AI revenues across those three companies are from OpenAI and Anthropic, two unprofitable, unsustainable companies that literally cannot afford to exist without these very same companies giving them money.

75
00:04:37.618 --> 00:04:41.140
Amazon sent $50 billion to OpenAI this year.

76
00:04:41.440 --> 00:04:43.300
They sent $5 billion to Anthropic.

77
00:04:43.600 --> 00:04:45.681
Google sent $10 billion to Anthropic.

78
00:04:46.221 --> 00:05:00.322
And in the next three and a half years, OpenAI and Anthropic, based on actual sell-side analyst evaluations, their estimates, that inform whether stock is going to go up or down after earnings, they are expecting $400 or more billion of revenue.

79
00:05:01.203 --> 00:05:07.368
30 or something percent of cloud growth just from these two unprofitable companies that will need to be given the money from somewhere.

80
00:05:08.049 --> 00:05:17.597
And on top of that, these companies have such low respect for the average investor, for the analysts, for everyone really, that they don't even disclose their AI revenues.

81
00:05:17.637 --> 00:05:19.699
The few times they deign us worthy...

82
00:05:20.119 --> 00:05:24.200
They use something called a run rate, an annualized run rate, which means, well, nothing.

83
00:05:24.220 --> 00:05:24.941
They never define it.

84
00:05:25.141 --> 00:05:26.661
It can mean month times 12.

85
00:05:27.061 --> 00:05:28.702
It can mean month times 13.

86
00:05:28.722 --> 00:05:30.462
It can mean last four weeks times 13.

87
00:05:30.883 --> 00:05:32.683
It's different every time, and they never define it.

88
00:05:32.803 --> 00:05:34.344
And then they sometimes just don't mention it.

89
00:05:34.804 --> 00:05:39.886
So you've got this big thing that is meant to be the biggest, most influential change to software ever.

90
00:05:40.406 --> 00:05:43.327
And whenever you ask them about it, when you say, how much are you making from this?

91
00:05:43.347 --> 00:05:44.767
They go, oh, I couldn't possibly say.

92
00:05:44.967 --> 00:05:45.688
I'm too shy.

93
00:05:46.108 --> 00:05:49.309
These are public companies, or at least the ones that aren't anthropic and open AI.

94
00:05:50.129 --> 00:05:51.390
When they have good news, they'll tell you.

95
00:05:51.450 --> 00:05:54.714
And when they don't tell you something, well, that actually speaks volumes.

96
00:05:55.434 --> 00:05:57.136
Have you used these tools?

97
00:05:57.176 --> 00:05:57.396
Yes.

98
00:05:57.436 --> 00:06:00.459
AI tools, Gemini, Anthropic, ChatGPT, etc.

99
00:06:01.020 --> 00:06:02.241
And you found no value in them?

100
00:06:02.721 --> 00:06:04.803
There's some value, but it's not.

101
00:06:04.944 --> 00:06:07.066
They have spent over a trillion dollars in capex.

102
00:06:07.486 --> 00:06:08.487
What does capex mean?

103
00:06:08.508 --> 00:06:09.349
Capital expenditures.

104
00:06:09.389 --> 00:06:15.257
So when you are a business and you have operating expenses like electricity, for example, those come right off immediately.

105
00:06:15.537 --> 00:06:18.721
Capital expenditures are long-term investments that are theoretically one-off.

106
00:06:18.741 --> 00:06:22.987
So a data center or indeed the GPUs you put inside an AI data center.

107
00:06:23.127 --> 00:06:27.890
Okay, so you've got a data center, and then you have these GPUs, which are like computer chips.

108
00:06:28.250 --> 00:06:32.513
So AI GPUs are much bigger, much more power intensive.

109
00:06:32.533 --> 00:06:34.235
They take a bunch of high bandwidth memory.

110
00:06:34.575 --> 00:06:42.500
And because of how many of them you need, you need thousands of them, tens of thousands, hundreds of thousands in some case, you need a bunch of power.

111
00:06:42.640 --> 00:06:48.064
So an example, OpenAI and Oracle are building a data center in Texas, in Abilene, Texas.

112
00:06:48.104 --> 00:06:50.586
1.2 gigawatts called Stargate Abilene.

113
00:06:51.186 --> 00:06:57.014
Within that, with each one of the eight buildings, there'll be 50,000 NVIDIA GB200 GPUs.

114
00:06:57.575 --> 00:07:00.799
So, City of Bristol takes about 780-800 megawatts of power a year, right?

115
00:07:00.819 --> 00:07:00.919
Yeah.

116
00:07:04.399 --> 00:07:12.023
Well, Stargate Abilene is condensing more power than that, 1.2 gigawatts, into a space around 1,172 times smaller.

117
00:07:12.583 --> 00:07:15.324
City of Bristol is about 1.2 billion square feet.

118
00:07:15.484 --> 00:07:18.005
Stargate Abilene is about 998,000.

119
00:07:18.566 --> 00:07:22.748
So you're condensing all of this power, all of this money, all of this labor into this one spot.

120
00:07:23.588 --> 00:07:26.391
And all of these data centers cost billions of dollars.

121
00:07:26.571 --> 00:07:29.374
All of these companies other than Microsoft are now to take out debt.

122
00:07:29.774 --> 00:07:34.159
And the thing is, they've spent over a trillion dollars so far, and they want to spend another trillion dollars next year.

123
00:07:34.559 --> 00:07:35.060
And for what?

124
00:07:35.580 --> 00:07:41.546
To make tens of billions of dollars, most of which comes from two unprofitable companies, Anthropic and OpenAI.

125
00:07:42.327 --> 00:07:50.073
One of the rebuttals to that would be that the adoption, the customer adoption of people using OpenAI and Anthropic has been absolutely insane.

126
00:07:50.134 --> 00:07:55.098
These are the fastest growing products in all of history, especially as it relates to sort of technology.

127
00:07:55.218 --> 00:08:07.088
If we just focus in on technology, there are, you know, hundreds and hundreds of millions of people, billions of people are using these tools every single day for things that they have subjectively decided are problems they need solving.

128
00:08:07.808 --> 00:08:10.789
So, you know, money is a lagging indicator of value.

129
00:08:11.429 --> 00:08:16.470
So one would argue that they're just investing ahead of the monetization options.

130
00:08:16.770 --> 00:08:18.150
The first, let's start with this adoption.

131
00:08:19.170 --> 00:08:23.591
Is it honest adoption when you are forced to use generative AI when you load Google?

132
00:08:23.671 --> 00:08:26.871
When you load Google Docs, Gemini screams in your ear.

133
00:08:27.232 --> 00:08:29.372
When you load Word, co-pilots bugging you.

134
00:08:29.772 --> 00:08:34.673
When you use Amazon, whatever Rufus AI is, has opinions on what socks you're buying.

135
00:08:35.533 --> 00:08:39.635
This is the largest non-consensual push of technology in history.

136
00:08:39.875 --> 00:08:40.916
ChatGPT, for example.

137
00:08:41.356 --> 00:08:44.978
Every single media outlet has been screaming about this for three years.

138
00:08:45.318 --> 00:08:46.919
They've been saying this will take your job.

139
00:08:47.199 --> 00:08:48.059
You must use this.

140
00:08:48.079 --> 00:08:50.140
If you don't use this, you're going to be falling behind.

141
00:08:50.641 --> 00:08:55.843
So people are using it because they've been told to use it constantly, and they're using it like search predominantly.

142
00:08:56.123 --> 00:08:59.025
And that's partly because Google fell behind search, and also because...

143
00:08:59.525 --> 00:09:01.026
It's better at ingesting queries sometimes.

144
00:09:01.046 --> 00:09:04.447
Sometimes if you use a generative search, it's like a trawling vessel.

145
00:09:04.487 --> 00:09:05.627
It's not very good at specifics.

146
00:09:05.647 --> 00:09:07.848
But if you're like, does this thing exist?

147
00:09:07.888 --> 00:09:09.449
Has this person ever said anything like this?

148
00:09:09.769 --> 00:09:13.010
It'll still probably get it wrong, but it'll scour the ocean for you.

149
00:09:13.330 --> 00:09:15.231
Nevertheless, that's not worth a trillion dollars.

150
00:09:15.691 --> 00:09:16.332
None of it is.

151
00:09:16.732 --> 00:09:19.053
The amount of money being sunk into this...

152
00:09:20.025 --> 00:09:21.786
is just incomparable to anything.

153
00:09:21.866 --> 00:09:28.808
Railways, it blows everything out of the water because there is no post-bubble story even for this.

154
00:09:29.349 --> 00:09:31.429
AI GPUs are not useful for other things either.

155
00:09:31.790 --> 00:09:45.475
It's a directionless egregore of capitalism, this headless beast that lumbers around, desperate to seek out growth everywhere, in the hopes that if it harasses people and scares people and...

156
00:09:46.095 --> 00:09:49.538
demonizes labor enough, people will be forced to use it.

157
00:09:50.399 --> 00:09:54.182
The reason I pause is because I just, I think about my own company.

158
00:09:54.202 --> 00:09:56.124
Obviously, everybody thinks about their own personal situation.

159
00:09:56.144 --> 00:10:06.032
So you have people listening now that don't use any AI tools, and you'll have people that are using it for everything, from coding, new software tools, to everything they write, to, you know, images, whatever.

160
00:10:06.333 --> 00:10:11.898
And when you look at the stats around enterprise adoption, it says 88% of organizations regularly use AI at least

161
00:10:12.658 --> 00:10:14.399
once for one particular business function.

162
00:10:14.779 --> 00:10:22.804
And I'd say in our company, 95% of people use one of these AI tools like Anthropic or ChatGPT or Gemini every day.

163
00:10:22.824 --> 00:10:23.425
Right.

164
00:10:23.525 --> 00:10:38.013
And that exists on some kind of spectrum of like the super users that are using it probably, you know, every hour of every day for almost everything to, you know, someone maybe hiring the executive team that's using it less because their job doesn't require of it as much.

165
00:10:38.574 --> 00:10:40.395
And when you look out into the world, you know,

166
00:10:41.495 --> 00:10:49.259
how the world is changing from a content perspective, if we're looking at generative AI, it is obvious that these tools are being widely adopted.

167
00:10:49.339 --> 00:10:52.281
Part of the symptom is the AI slop you see all over the internet.

168
00:10:53.201 --> 00:10:59.365
So this idea that it's not being used, I struggle with.

169
00:10:59.705 --> 00:11:00.605
It's being used.

170
00:11:00.805 --> 00:11:01.826
Here's the thing with the slop.

171
00:11:02.326 --> 00:11:09.770
Before we had AI slop, we had SEO slop because Google incentivized doing the lowest common denominator that would rank well in search.

172
00:11:10.110 --> 00:11:18.715
It's a whole story about how they pulled back spam guards, thanks to Prabhagar Raghavan, which we can get into, where they made the internet worse by allowing worse content to rank higher.

173
00:11:19.095 --> 00:11:27.279
It's why we had, when you used to Google our best washing machine, there's 11 different horrible blogs that read like somebody got a concussion.

174
00:11:27.599 --> 00:11:32.022
They are built to rank rather than be read by humans or built to be made good.

175
00:11:32.842 --> 00:11:35.003
So AI helps weaponize that scale.

176
00:11:35.023 --> 00:11:37.363
Yeah, you can make a bunch of generic slop.

177
00:11:37.583 --> 00:11:38.944
We've had slop for years.

178
00:11:39.164 --> 00:11:40.564
We've just found a slop machine.

179
00:11:41.144 --> 00:11:43.045
But then also there's the problem of cost.

180
00:11:43.365 --> 00:11:45.486
So when you use AI services, you burn tokens.

181
00:11:45.986 --> 00:11:47.286
And it's per million tokens.

182
00:11:47.366 --> 00:11:48.026
What's a token?

183
00:11:48.326 --> 00:11:50.047
So it's around three quarters of a word.

184
00:11:50.727 --> 00:11:52.128
So it's characters.

185
00:11:52.268 --> 00:11:56.812
So the AI companies have a currency in which they charge you, like a taxi in New York has a meter.

186
00:11:57.192 --> 00:11:57.392
Yeah.

187
00:11:57.892 --> 00:11:59.654
And they call it tokens.

188
00:11:59.714 --> 00:11:59.894
Yeah.

189
00:12:00.554 --> 00:12:02.796
And every word, let's just say for ease it's a word.

190
00:12:02.896 --> 00:12:04.397
Yeah, about a word.

191
00:12:04.437 --> 00:12:04.657
Yeah.

192
00:12:04.837 --> 00:12:06.238
And it's per million tokens.

193
00:12:06.298 --> 00:12:09.741
So you'll be charged per million input tokens, the stuff you feed into it.

194
00:12:09.861 --> 00:12:12.283
like a document or a code base.

195
00:12:12.583 --> 00:12:16.886
And the output tokens are both the stuff it spits out at the end, but also when it thinks.

196
00:12:17.106 --> 00:12:20.708
So, okay, you've asked me to give you the best restaurants in this area of New York.

197
00:12:21.008 --> 00:12:22.469
I should find the best restaurants in New York.

198
00:12:22.709 --> 00:12:24.030
All of that's output tokens as well.

199
00:12:24.651 --> 00:12:28.213
However, when you're paying for a monthly service, you don't see any of that.

200
00:12:28.753 --> 00:12:30.014
Put all that crap to the side.

201
00:12:30.394 --> 00:12:31.355
They just have rate limits.

202
00:12:31.375 --> 00:12:35.277
So you can use them a certain amount, and then when you run out, but they kind of obfuscate what that was.

203
00:12:35.497 --> 00:12:35.678
Now...

204
00:12:36.438 --> 00:12:39.623
Someone recently found, Semi Analysis actually found this, a big analyst group.

205
00:12:40.164 --> 00:12:47.035
They found that on a $200 a month chat GPT subscription, you can burn $14,000 worth of tokens.

206
00:12:47.977 --> 00:12:49.860
And on Anthropix, you can burn $8,000 for...

207
00:12:52.053 --> 00:13:13.058
200 bucks that is how most and even on the 20 buck a month service you can burn 400 dollars now most people don't realize that most people have no idea what ai costs most people just think oh it's 20 bucks a month no all of these companies run a horrifying loss open ai lost 20.9 billion dollars last year because people can burn as many tokens as they want

208
00:13:13.538 --> 00:13:26.313
And when they tried to move everybody on the enterprise side, so companies bigger than 150, onto actually paying the cost of AI in around March of 2026, to quote Sam Allman, they said, people have a big problem with it, I think.

209
00:13:26.373 --> 00:13:29.116
It's a huge issue, which is not really what the...

210
00:13:29.777 --> 00:13:53.446
heir apparent to tech's history is meant to be saying but the point is enterprises immediately started freaking out uber burned through their entire annual token budget in three months so suddenly after everyone's saying ai is the most productive thing ever it's amazing it's changing everything the moment people actually had to pay for it they go oh i don't know actually um maybe it's it's obviously we all love it it's all great right

211
00:13:54.624 --> 00:13:56.987
But it's costing too much, so we now need to reduce the cost.

212
00:13:57.007 --> 00:14:00.250
Because people are just dumping stuff into it, being like, what do I do here?

213
00:14:00.450 --> 00:14:02.652
And getting whatever the median is out.

214
00:14:02.672 --> 00:14:04.074
Because that's what these things do.

215
00:14:04.154 --> 00:14:05.555
They provide the median answer.

216
00:14:06.576 --> 00:14:09.439
So essentially, someone like me who's a power user of these tools...

217
00:14:10.060 --> 00:14:17.362
I could be costing Anthropic or OpenAI $1,000, but they're only charging me $100, let's say.

218
00:14:17.822 --> 00:14:24.364
So they are having to subsidize $900 of my usage because of the electricity costs and the costs at their data centers.

219
00:14:24.824 --> 00:14:27.165
And so your assertion here is that that is unsustainable.

220
00:14:27.365 --> 00:14:27.605
Yes.

221
00:14:27.785 --> 00:14:30.528
And just to be clear, they're probably not one for one dollar.

222
00:14:30.548 --> 00:14:31.549
It might be 34.

223
00:14:31.609 --> 00:14:32.430
We don't we don't know.

224
00:14:32.670 --> 00:14:33.671
I think it's unprofitable.

225
00:14:34.131 --> 00:14:35.313
These companies don't disclose them.

226
00:14:35.413 --> 00:14:38.716
Even in their auditive financials, they play funny games with how they categorize things.

227
00:14:38.876 --> 00:14:40.077
But nevertheless, yes.

228
00:14:40.878 --> 00:14:46.783
And on top of that, the way that you stand up inference, which is the thing that creates the output within these data centers, is...

229
00:14:47.624 --> 00:14:50.686
You're not just saying, okay, turn the inference machine on, let's go.

230
00:14:51.206 --> 00:14:58.090
You are standing up the GPUs necessary to take in the demand, and if you buy too much, you've wasted the money.

231
00:14:58.931 --> 00:15:01.472
You have to pay for the hourly GPU use regardless.

232
00:15:01.932 --> 00:15:04.034
If you buy too few, your customers can't use it.

233
00:15:04.054 --> 00:15:06.055
They get pissed off at you, they can, so they go with someone else.

234
00:15:06.095 --> 00:15:10.858
But nevertheless, yeah, they would get demand selling $20 or $40 for a dollar.

235
00:15:11.238 --> 00:15:12.399
And that's what these services do.

236
00:15:12.759 --> 00:15:14.380
And really, the simplest way to explain it is,

237
00:15:14.740 --> 00:15:21.848
If they were actually profitable, if they believed that these services were worthwhile and that they were worthy of the cost, they'd charge it.

238
00:15:22.209 --> 00:15:24.631
Regular people wouldn't be able to get a monthly subscription.

239
00:15:25.152 --> 00:15:27.334
They'd just be paying what it's worth.

240
00:15:27.515 --> 00:15:30.098
Unless, of course, there was an economic problem.

241
00:15:30.278 --> 00:15:31.079
And it's very simple.

242
00:15:32.134 --> 00:15:35.978
You pay when you use an LLM, regardless of whether you get what you want.

243
00:15:36.339 --> 00:15:44.588
When these things hallucinate, say you're doing something, you're coding something, and they go through a code base and they fuck up a bunch of stuff, they break a bunch of stuff, you're paying for that.

244
00:15:44.728 --> 00:15:46.110
You're paying for it whether it works or not.

245
00:15:46.410 --> 00:15:48.192
Unless, of course, you're using one of these subscriptions.

246
00:15:48.868 --> 00:16:07.234
I think the really interesting point is, are they spending ahead of the value showing up, which is, I imagine, what they would argue, or are they spending all of this money and subsidizing all of their users in a way that's unsustainable and that will never be justified?

247
00:16:07.594 --> 00:16:14.996
Because you think back through the history of technology, you often get people losing money to grab market share.

248
00:16:15.216 --> 00:16:18.137
And they're also focusing on bringing the costs down

249
00:16:19.395 --> 00:16:21.357
and making it more profitable for them as well.

250
00:16:21.777 --> 00:16:23.679
But they can't afford to underinvest.

251
00:16:24.360 --> 00:16:28.403
If they were bringing the cost down, they would have brought the cost down, which they have not.

252
00:16:28.904 --> 00:16:30.525
It seems to be getting more expensive.

253
00:16:30.565 --> 00:16:33.528
In fact, everyone, inference providers, don't seem to be profitable.

254
00:16:33.888 --> 00:16:36.531
Even the companies renting out GPUs don't seem to be profitable.

255
00:16:37.191 --> 00:16:41.495
I imagine that it wasn't like they started out and they were like, shit, this is unprofitable.

256
00:16:41.515 --> 00:16:42.316
At the beginning, we know it.

257
00:16:42.556 --> 00:16:42.917
Screw it.

258
00:16:43.237 --> 00:16:44.378
We'll keep doing it.

259
00:16:44.418 --> 00:16:45.019
I don't think it's...

260
00:16:45.479 --> 00:16:56.486
some big conspiracy, they probably thought at some point, yeah, this will go profitable, the chips will catch up, customers will pay for the overwhelming value, because you don't know in 2023 where it's going to be in 2026.

261
00:16:56.506 --> 00:16:57.667
You assume it's going to go up.

262
00:16:57.687 --> 00:16:59.008
That's the nature of venture capital.

263
00:16:59.568 --> 00:17:03.971
They should have stopped in 2024 when OpenAI lost over $5 billion.

264
00:17:04.031 --> 00:17:06.212
They should have been like, yep, this is not going to work.

265
00:17:06.753 --> 00:17:09.995
But they kept going because it helped number go up so much.

266
00:17:10.155 --> 00:17:11.656
It helped stock values pump.

267
00:17:11.736 --> 00:17:12.937
It helped everyone pump.

268
00:17:12.997 --> 00:17:15.018
It helped NVIDIA pump, Microsoft, everyone.

269
00:17:15.518 --> 00:17:17.099
And not from the revenues.

270
00:17:17.300 --> 00:17:20.442
Because here's the funny thing about Google, Microsoft, and Amazon.

271
00:17:20.622 --> 00:17:24.084
People, for years, have been saying their AI bets have paid off.

272
00:17:24.424 --> 00:17:25.685
Wow, their AI bets have paid off.

273
00:17:25.925 --> 00:17:28.467
As these companies refuse to say how much they're making from AI.

274
00:17:28.907 --> 00:17:36.052
But because their existing businesses continue to grow, and did so, by the way, through price increases, changes to how Google and Meta...

275
00:17:36.932 --> 00:17:37.993
did advertising.

276
00:17:38.273 --> 00:17:40.595
Amazon bumped up prices and changed how they did.

277
00:17:40.635 --> 00:17:44.097
Actually, Amazon started a remarkable ad business during this whole time as well.

278
00:17:44.397 --> 00:17:45.678
And they're selling through Amazon platform.

279
00:17:45.698 --> 00:17:55.064
Anyway, nothing to do with AI, but because number go up, because revenue go up, everyone went, it's AI because these companies wouldn't spend a trillion dollars for no reason, right?

280
00:17:55.625 --> 00:17:59.047
Except in fiscal year 2026, which just ended for Microsoft.

281
00:17:59.367 --> 00:17:59.968
Annoying, I know.

282
00:18:00.728 --> 00:18:05.611
They made, total according to Bloomberg, about $34.33 billion.

283
00:18:05.651 --> 00:18:08.132
$24.1 billion of that was from OpenAI.

284
00:18:08.873 --> 00:18:15.237
So that leaves them with about $10 billion in a year when they spent $115 billion on capital expenditures and intend to spend $175 billion next year.

285
00:18:17.738 --> 00:18:19.360
The math does not make sense.

286
00:18:19.720 --> 00:18:26.666
I imagine their plan was, okay, this is just going to get exponentially more valuable, and at some point, the costs will be outpaced by the return.

287
00:18:27.386 --> 00:18:30.449
Problem is that large language models need a bunch of money to train them.

288
00:18:30.649 --> 00:18:32.190
They need constant data flows through it.

289
00:18:32.210 --> 00:18:33.371
They need customized data.

290
00:18:33.692 --> 00:18:36.014
It's just this big, expensive monster.

291
00:18:36.554 --> 00:18:41.979
And when you try and talk to people about it, and you try and say, hey, look, this is really bad.

292
00:18:42.039 --> 00:18:42.719
NVIDIA has...

293
00:18:43.580 --> 00:18:49.025
sold, so it was $215.9 billion in the last fiscal year worth of GPUs mostly.

294
00:18:49.906 --> 00:19:02.897
And you try and go, yeah, that's the support, like $22 billion of revenue total in the entire world outside of these two companies that literally require money being fed into them, sometimes by NVIDIA to keep alive.

295
00:19:02.917 --> 00:19:03.157
Yeah.

296
00:19:04.142 --> 00:19:07.044
When you tell people that, they go, well, companies just lose money, right?

297
00:19:07.184 --> 00:19:15.729
Because we have this, quote Ed Elson from Prof G Markets, we have this cult-like worship of the wealthy, where we think that someone wouldn't spend all this money for no reason, right?

298
00:19:15.930 --> 00:19:19.332
Because reconciling with that, with this idea that

299
00:19:20.548 --> 00:19:24.989
The ultra wealthy, the ultra powerful didn't get there through big brains.

300
00:19:25.049 --> 00:19:36.851
They didn't get there through anything other than luck and opportunism and getting an MBA perhaps with the right people, that they just got there because they're regular people and they just happened to be in the right place at the right time.

301
00:19:36.911 --> 00:19:41.912
Reconciling with that and realizing that the world is not controlled by people like a meritocracy is kind of grim.

302
00:19:42.092 --> 00:19:44.872
So it's easy to be like, no, they're not making a mistake.

303
00:19:45.293 --> 00:19:46.393
I must be missing something.

304
00:19:46.593 --> 00:19:47.373
And that's what they want.

305
00:19:48.073 --> 00:19:54.137
So, you know, I think back through the history of technological breakthroughs, and I think about, I mean, you can look at different industries.

306
00:19:54.177 --> 00:19:57.019
And one of my favorite books on this subject is The Innovator's Dilemma.

307
00:19:57.279 --> 00:19:57.579
I read it.

308
00:19:57.999 --> 00:20:03.563
And one of the things it talks about is how the innovation that ends up taking out or transforming an industry...

309
00:20:04.992 --> 00:20:09.134
often starts worse, doesn't make economic sense.

310
00:20:09.354 --> 00:20:10.775
None of your customers are asking for it.

311
00:20:11.075 --> 00:20:12.896
And this is typically why we end up ignoring it.

312
00:20:12.936 --> 00:20:15.497
So like, you've got horse and carriages in the 1800s.

313
00:20:16.137 --> 00:20:18.418
Amazing form of transport, according to the 1800s.

314
00:20:18.518 --> 00:20:19.579
You know, people of the 1800s.

315
00:20:19.859 --> 00:20:21.399
And then you have this thing called cars come along.

316
00:20:21.840 --> 00:20:23.700
Now, the problem with cars is they broke down all the time.

317
00:20:23.921 --> 00:20:25.381
It's kind of like AI hallucinates now.

318
00:20:26.041 --> 00:20:28.803
They were more expensive, and the economics of it didn't make sense.

319
00:20:28.823 --> 00:20:30.443
You might as well walk, then buy a car.

320
00:20:30.744 --> 00:20:33.525
There was a law at the time that meant you had to walk in front of it with a red flag.

321
00:20:34.065 --> 00:20:37.287
and someone had to employ someone to walk in front of it waving a red flag.

322
00:20:37.727 --> 00:20:38.627
Obviously, it's worse.

323
00:20:38.647 --> 00:20:39.708
It's a worse solution.

324
00:20:40.128 --> 00:20:48.512
However, these things that are disruptive innovations, they have a higher ceiling of growth, and so they eventually overtake the horse.

325
00:20:49.272 --> 00:20:54.616
And when I think about that analogy in the context of all of this, I go, okay, it's imperfect at the moment.

326
00:20:55.137 --> 00:20:57.939
The economic models aren't perfectly ironed out.

327
00:20:58.199 --> 00:21:01.402
They're still figuring out how to make it cheaper, the infrastructure, etc.

328
00:21:01.962 --> 00:21:05.385
But if you think about the rate of improvement versus other, you know...

329
00:21:06.070 --> 00:21:12.673
let's say coding, how much could I train a human coder to improve and to increase their output versus an AI agent?

330
00:21:13.153 --> 00:21:28.139
One would go, if you just imagine any rate of improvement in these AI tools, at some point, if you just imagine a 5% rate of improvement per month, at some point it's, you know, and then you imagine a 5% reduction in cost, which is what we did with the internet, what we did with cars.

331
00:21:28.299 --> 00:21:29.360
Yeah, but that's- Moore's law.

332
00:21:29.380 --> 00:21:31.701
Moore's law is a theory, and Moore's law is not with GPUs.

333
00:21:32.261 --> 00:21:33.561
So let me actually explain.

334
00:21:33.581 --> 00:21:34.482
So, NVIDIA.

335
00:21:35.162 --> 00:21:43.406
NVIDIA invented, I think it was in the 2000s, they put out something called CUDA, which is the underlying software library and the way to run software on GPUs.

336
00:21:43.666 --> 00:21:50.369
It took them a solid decade or more to make it something where they could do data analytics, one of the early things, mapper and such.

337
00:21:50.810 --> 00:21:53.111
And then when AI came along, they'd had lots of experience with it.

338
00:21:53.191 --> 00:21:57.393
But nevertheless, this company has got more money, more attention.

339
00:21:58.193 --> 00:22:04.639
more geniuses behind them, more people focused on making their things more efficient than anyone could ever ask for.

340
00:22:04.879 --> 00:22:07.201
And NVIDIA, for anyone that doesn't know, makes the chips.

341
00:22:07.321 --> 00:22:12.446
So, and that CUDA thing I mentioned, they were the ones with CUDA, and CUDA allowed generative AI to grow.

342
00:22:12.706 --> 00:22:13.787
Okay, so they're chips.

343
00:22:14.027 --> 00:22:14.507
Chips, yes.

344
00:22:14.527 --> 00:22:15.388
And chips are needed.

345
00:22:15.428 --> 00:22:16.849
Those are the things that go into the data centers.

346
00:22:16.889 --> 00:22:20.212
And their specific chips are the ones where you can run AI software on it.

347
00:22:20.292 --> 00:22:22.034
So the training runs and also the inference.

348
00:22:22.234 --> 00:22:23.375
Now, here's the thing.

349
00:22:24.196 --> 00:22:25.036
The car example.

350
00:22:25.797 --> 00:22:30.863
Back then, you didn't have pretty much every mathematician and scientist going into the car industry.

351
00:22:31.143 --> 00:22:34.226
You didn't have the combined world's governments never shutting up about this.

352
00:22:34.507 --> 00:22:37.690
And by the way, giving them credit early.

353
00:22:38.291 --> 00:22:40.614
Since 2023, they've been saying this is inevitable.

354
00:22:40.934 --> 00:22:42.676
Even what you said, 5% improvement.

355
00:22:43.176 --> 00:22:45.960
I don't even know how you'd measure that because a junior software engineer,

356
00:22:46.500 --> 00:22:51.582
can still experience things and learn things from context, from how people deal with problems.

357
00:22:52.022 --> 00:22:56.384
And the way that people deal with problems is not as simple as looking at the code or reading some emails.

358
00:22:56.884 --> 00:22:58.564
It's context cues from speaking to a person.

359
00:22:58.884 --> 00:23:00.845
It's being in different environments.

360
00:23:01.285 --> 00:23:03.706
And there are uses for LLMs in coding.

361
00:23:03.946 --> 00:23:05.207
I don't dispute that.

362
00:23:05.787 --> 00:23:08.268
But even saying 5%, what does that mean?

363
00:23:08.368 --> 00:23:09.328
Is it better at Rust?

364
00:23:09.368 --> 00:23:10.248
Is it better at C++?

365
00:23:10.308 --> 00:23:11.569
I'd say productivity.

366
00:23:11.789 --> 00:23:12.989
So just like, yeah, shipped.

367
00:23:13.049 --> 00:23:16.150
If we did it in the context of coding, it would be like shipped code.

368
00:23:16.270 --> 00:23:16.810
That's the thing.

369
00:23:16.830 --> 00:23:20.611
That would be like, he's the best writer in the world because his newsletter is really long.

370
00:23:20.791 --> 00:23:22.612
That's an insane way of evaluating it.

371
00:23:22.792 --> 00:23:30.613
With coding, it would be, I mean, it's even difficult to evaluate because it's, is the software out there better is actually a great way of evaluating it.

372
00:23:30.833 --> 00:23:32.694
And I would say uniformly not.

373
00:23:32.914 --> 00:23:40.356
I would say the standard of software across Google, Microsoft, Amazon, Meta, especially God, Meta's a monstrosity, is worse.

374
00:23:41.796 --> 00:23:48.461
GitHub, someone posted on Twitter earlier today, we should get a notification when GitHub is up rather than when it's down because that would be more reliable.

375
00:23:48.802 --> 00:23:52.464
Microsoft is one of the largest companies in the world and they can barely wipe their own ass when it comes to GitHub.

376
00:23:53.285 --> 00:24:03.173
The quality of software is going down, weirdly enough, as more people use LLMs and more businesses demand, and I really do mean demand, that people use these services.

377
00:24:03.453 --> 00:24:13.936
So on this point of, if we go back to this horse and carriage and car analogy, say that we're at whatever point today, if you imagine any rate of improvement in the technology, which we have seen since Chatjipati came out.

378
00:24:14.497 --> 00:24:18.838
I remember when Chatjipati came out and I was in Asia, and I was there showing it to my fiancé, and I was like, look, he can do this.

379
00:24:19.118 --> 00:24:21.419
And it was hallucinating once in a while and getting things wrong.

380
00:24:21.819 --> 00:24:24.340
I actually don't have that experience anymore.

381
00:24:24.800 --> 00:24:27.941
I have moments where I believe its reasoning is weak,

382
00:24:28.621 --> 00:24:31.483
but I don't have outright hallucinations anymore.

383
00:24:31.863 --> 00:24:34.144
See, I disagree.

384
00:24:34.184 --> 00:24:36.165
Give me an example of what you define as a hallucination.

385
00:24:36.185 --> 00:24:36.925
Okay, great one.

386
00:24:36.965 --> 00:24:38.306
So I have a Bloomberg terminal.

387
00:24:38.386 --> 00:24:40.427
The very useful thing they have on there is Ask B.

388
00:24:40.607 --> 00:24:49.452
So when you do a Bloomberg inquiry to look up what we think NVIDIA's revenue is going to be next quarter, it runs something called BQL, which is its own programming language.

389
00:24:49.832 --> 00:24:55.033
Now, instead of having to learn that, you can just type into Ask B and it will generate it and run it for you.

390
00:24:55.113 --> 00:24:57.434
And so you get pulled up and you know where the date is coming from.

391
00:24:57.454 --> 00:24:58.954
It deals with hallucinations real well.

392
00:24:59.414 --> 00:25:00.774
The other day, I was like, you know what?

393
00:25:00.994 --> 00:25:01.794
We'll get a little spicy.

394
00:25:01.814 --> 00:25:09.416
I'm going to look up the growth rate of stocks of Microsoft, Google, Meta, and Amazon over the course of five years, I think it was.

395
00:25:10.296 --> 00:25:14.197
And I was about to copy-paste it over to something, looked at it in Excel.

396
00:25:14.257 --> 00:25:15.517
It was right in the newsletter.

397
00:25:15.537 --> 00:25:17.938
I went, Microsoft stock's never been $575 a stock.

398
00:25:21.039 --> 00:25:21.600
You know what?

399
00:25:21.780 --> 00:25:24.643
When it's a cute little thing like, oh, it's a stock price and I can't afford it.

400
00:25:24.663 --> 00:25:25.644
It was no harm, no foul.

401
00:25:25.844 --> 00:25:26.384
That's fine.

402
00:25:27.385 --> 00:25:34.552
But when you're talking about a transcribing tool for a doctor or a financial model that a hedge fund is dependent on.

403
00:25:35.699 --> 00:25:38.202
At that point, it becomes a little more dangerous.

404
00:25:39.082 --> 00:25:47.130
And the thing is, a hallucination with a software package, for example, refactoring a code base, and it leaves the door open security-wise.

405
00:25:47.451 --> 00:25:48.892
Or it just breaks something.

406
00:25:49.172 --> 00:25:52.115
And you, I don't know, maybe you've been vibe coding for six months.

407
00:25:52.135 --> 00:25:54.317
You haven't really been coding with your own hands for a while.

408
00:25:54.337 --> 00:25:55.438
Maybe you've forgotten a few things.

409
00:25:55.879 --> 00:25:57.300
You have this slot to look for.

410
00:25:57.320 --> 00:25:58.181
You go, fuck, you know what?

411
00:25:58.201 --> 00:25:58.682
I'm not doing it.

412
00:25:59.813 --> 00:26:01.714
And so the problems become multiplicative.

413
00:26:02.394 --> 00:26:06.797
And I don't really know how you train them out of that, and they've certainly not succeeded.

414
00:26:07.217 --> 00:26:10.079
So on one hand, they have got better.

415
00:26:10.699 --> 00:26:16.843
But one of the main ways they evaluate them getting better are benchmarks that are adjusted specifically for large language models.

416
00:26:17.183 --> 00:26:18.724
Because you can't just have them do...

417
00:26:19.324 --> 00:26:34.682
tasks they've got better at that they found some tasks they can have them do on but even them like meter metr they have this thing where it's like check out this chart look how much better it's getting at running tasks wow it can go for an hour and then you look it's like yeah and successfully completing them 50 of the time they have a hallucination

418
00:26:35.022 --> 00:26:35.422
leaderboard.

419
00:26:35.742 --> 00:26:37.883
And it really focuses on basic tasks.

420
00:26:37.903 --> 00:26:58.671
And it shows that the four-year trend, according to historical data from the Victoria Hallucination Leaderboard, shows that hallucination rates on simple summarization tasks have plummeted from around 21%, 21.8% four years ago, down to 0.7% roughly on today's top frontier models like Gemini and ChatGPT.

421
00:26:59.131 --> 00:27:02.512
Again, the point of nuance here is that these are on simple tasks, right?

422
00:27:02.592 --> 00:27:03.893
which is kind of what I've experienced.

423
00:27:03.913 --> 00:27:06.134
I've experienced that on day-to-day things that hallucinates less.

424
00:27:06.514 --> 00:27:07.794
Again, rate of improvement thinking.

425
00:27:07.854 --> 00:27:16.938
So if I just imagine the trajectory to continue, there is going to become a time where hallucinations become rarer than they are today, increasingly.

426
00:27:17.378 --> 00:27:20.480
And also what I would say is when I think about other technologies, there's two more points.

427
00:27:21.140 --> 00:27:26.505
Other technologies at their inception, when they first came to the world, like the internet, also had technical difficulties.

428
00:27:27.066 --> 00:27:31.930
I remember growing up with dial-up modems, and I couldn't go on the phone at the same time as going on the internet.

429
00:27:31.970 --> 00:27:34.613
I'd have to stop RuneScape upstairs to go on the phone.

430
00:27:35.073 --> 00:27:37.115
And you thought, this is crap, this technology is crap.

431
00:27:37.616 --> 00:27:38.276
I don't know, mate.

432
00:27:38.316 --> 00:27:38.737
I loved it.

433
00:27:38.757 --> 00:27:38.877
Yeah.

434
00:27:38.997 --> 00:27:39.558
Yeah, I know.

435
00:27:39.618 --> 00:27:40.899
It felt like magic.

436
00:27:41.059 --> 00:27:45.663
And then, in hindsight, you go, wow, I now have Starlink and 5G internet from my phone.

437
00:27:46.124 --> 00:27:47.165
It's unbelievable.

438
00:27:47.485 --> 00:27:49.066
You couldn't leave the house with internet before.

439
00:27:49.567 --> 00:27:51.248
And that's what I mean by the rate of improvement thinking.

440
00:27:51.429 --> 00:27:54.591
I'd say the last point is we often compare...

441
00:27:55.732 --> 00:27:57.353
AI to perfection.

442
00:27:57.713 --> 00:27:57.953
Right.

443
00:27:57.973 --> 00:27:59.934
Whereas that's not actually the alternative.

444
00:28:00.715 --> 00:28:13.161
In the working world, like if I wanted to do, let's say, a simple writing task, I should compare AI to my alternative way of doing that simple writing task, which is both measured in my time, right?

445
00:28:13.281 --> 00:28:15.902
And my ability to hallucinate as a person who doesn't know everything.

446
00:28:17.507 --> 00:28:23.512
Or if I'm hiring someone, an intern who might also be prone to hallucination or have gaps in their knowledge.

447
00:28:24.132 --> 00:28:26.974
So it's not actually like we're comparing, we should compare AI to perfection.

448
00:28:26.995 --> 00:28:28.516
It's AI to the other alternatives.

449
00:28:28.956 --> 00:28:37.322
And if someone hallucinates 0.7% of the time, but knows way more and is faster, maybe on a net basis, that's a good trade.

450
00:28:37.703 --> 00:28:38.563
Maybe I should use AI.

451
00:28:38.824 --> 00:28:40.845
So let's start with an example.

452
00:28:41.285 --> 00:28:42.026
Someone I love dearly.

453
00:28:42.106 --> 00:28:43.047
Matt Hughes, my editor.

454
00:28:43.267 --> 00:28:43.467
Yeah.

455
00:28:43.487 --> 00:28:44.468
He lives inside of Liverpool.

456
00:28:44.908 --> 00:28:45.369
Wonderful guy.

457
00:28:45.389 --> 00:28:45.489
Yeah.

458
00:28:46.129 --> 00:28:49.032
I don't pay Matt Hughes because he knows everything.

459
00:28:49.272 --> 00:28:52.715
I pay him because he has incredible context and a ton of knowledge.

460
00:28:53.135 --> 00:28:55.217
And he's willing to expand it and work with me.

461
00:28:55.637 --> 00:28:56.718
And moral sport.

462
00:28:56.758 --> 00:28:57.839
And he's a great editor.

463
00:28:57.859 --> 00:29:01.702
But he's also someone who gets into the guts of it and has the experiences of it.

464
00:29:01.742 --> 00:29:02.003
He's a...

465
00:29:02.603 --> 00:29:17.892
decorated tech journalist and on top of that a wonderful loving being with empathy and joy in his heart for the stuff he loves and absolute fucking venom for the people he hates that's i can't get that from a large language model but on top of that i don't i push back on just the assumption there

466
00:29:18.692 --> 00:29:23.315
When you say knows everything, what good is something that knows everything when it sometimes doesn't know anything?

467
00:29:23.455 --> 00:29:24.256
When it's sometimes.

468
00:29:24.516 --> 00:29:28.179
And the thing is, are you really paying an intern for something basic?

469
00:29:28.479 --> 00:29:32.041
Are you really going to them and saying, yeah, can you look up what the date is?

470
00:29:32.121 --> 00:29:33.142
No, you're doing that on Google.

471
00:29:33.602 --> 00:29:37.385
Whatever the task is, you are trying to also train an intern.

472
00:29:37.545 --> 00:29:40.967
The point of an intern is to train them and turn them in, take them out of Pinocchio status.

473
00:29:40.987 --> 00:29:41.087
Yeah.

474
00:29:41.908 --> 00:29:43.709
But it's also, an intern learns.

475
00:29:43.749 --> 00:29:44.790
An intern gets context.

476
00:29:44.850 --> 00:29:46.070
An intern learns your habits.

477
00:29:46.170 --> 00:29:47.751
An AI gets context and learns.

478
00:29:47.791 --> 00:29:48.371
No, it doesn't.

479
00:29:48.732 --> 00:29:49.492
An AI doesn't learn?

480
00:29:49.672 --> 00:29:50.513
I mean, it doesn't.

481
00:29:50.773 --> 00:29:56.496
The way it learns is you create a giant clawed.md file that it sometimes doesn't read, sometimes does read.

482
00:29:56.716 --> 00:29:57.916
You create a harness.

483
00:29:58.157 --> 00:30:01.298
It's like it's Pee-wee's breakfast machine from Pee-wee's playhouse.

484
00:30:01.318 --> 00:30:05.200
You have to do all these contrivances to mitigate the hallucinations.

485
00:30:05.220 --> 00:30:06.181
And even then at the end...

486
00:30:06.661 --> 00:30:07.922
How much effort have you put in?

487
00:30:08.122 --> 00:30:10.603
But, okay, this is an extreme simplified example.

488
00:30:10.763 --> 00:30:13.644
If I went on my claw down and said, what's my dog's name?

489
00:30:13.825 --> 00:30:14.625
It would know my dog's name.

490
00:30:15.005 --> 00:30:16.066
Jesus Christ.

491
00:30:16.286 --> 00:30:18.947
This company raised $95 billion this year.

492
00:30:19.027 --> 00:30:22.949
I'm using an extreme simplified example to show that it can remember things from the past.

493
00:30:23.349 --> 00:30:25.791
Obviously, it knows much more complex things as well.

494
00:30:26.091 --> 00:30:27.011
But I just use that as an example.

495
00:30:27.231 --> 00:30:30.673
So we accept the fact that it does have memory of the past.

496
00:30:30.953 --> 00:30:35.575
It has files it can access that have stuff on it, but that's not the same as memory.

497
00:30:35.955 --> 00:30:38.397
And it's also just, okay, so it remembers your dog's name.

498
00:30:38.737 --> 00:30:40.297
It might remember your habits.

499
00:30:40.418 --> 00:30:42.138
It might be able to read things you've said before.

500
00:30:42.899 --> 00:30:44.239
Does it know your moods?

501
00:30:44.319 --> 00:30:46.640
Does it know what's going on in the world around it?

502
00:30:46.801 --> 00:30:47.921
Does it have good days and bad days?

503
00:30:48.121 --> 00:30:48.962
Is it there for you?

504
00:30:49.222 --> 00:30:50.983
Because it's just a fucking text machine.

505
00:30:51.343 --> 00:30:53.504
And the thing is, the intent example.

506
00:30:54.224 --> 00:30:56.065
An intern is something that can grow.

507
00:30:56.205 --> 00:30:57.606
It's something that you invest in.

508
00:30:57.887 --> 00:31:00.428
That's not something you do through feeding files and text to it.

509
00:31:00.709 --> 00:31:10.336
The way that we store memories ourselves, the way in which we accrue experiences is a milestone of emotion and feelings and facts.

510
00:31:10.356 --> 00:31:10.916
Completely different.

511
00:31:11.156 --> 00:31:12.878
So I think there's two things here.

512
00:31:12.898 --> 00:31:16.361
There's the process in which something happens, and then there's the output.

513
00:31:17.241 --> 00:31:20.244
So the process, you were describing the process of how a human does memory.

514
00:31:20.864 --> 00:31:27.250
The way that an AI does memory is different, but the thing that people care about is their value in the output.

515
00:31:27.570 --> 00:31:33.656
I.e., you know, if I dump all of my files into Claude, I don't really care how it processes it, as long as when I ask it...

516
00:31:34.376 --> 00:31:35.397
what's my revenue?

517
00:31:35.757 --> 00:31:36.458
It has the number.

518
00:31:36.618 --> 00:31:38.460
And one could say the same thing about training someone.

519
00:31:38.480 --> 00:31:47.348
You could say you teach them, you put lots of effort into them, you give them lots of context, you educate them and give them experiences, and then you might come and say to them, by the way, what's my revenue?

520
00:31:47.748 --> 00:31:50.631
Now, the processes are entirely different, but the outcome is what I care about.

521
00:31:51.031 --> 00:31:52.472
Do they know the revenue number when I ask them?

522
00:31:52.492 --> 00:31:52.673
Yeah.

523
00:31:53.273 --> 00:31:55.514
And so I think that's the part that we sometimes get lost in.

524
00:31:55.534 --> 00:31:58.816
We get, you know, because I've heard this debate about, like, can AI be creative?

525
00:31:59.336 --> 00:31:59.616
Right.

526
00:31:59.716 --> 00:32:03.278
I think, like, the way to answer that question is, like, it's about the output.

527
00:32:03.418 --> 00:32:06.360
When I ask it to do a creative thing, does it give me the answer?

528
00:32:06.580 --> 00:32:09.141
Not, is the process the same as a human process?

529
00:32:09.822 --> 00:32:11.483
Because actually, no, who cares what the product?

530
00:32:11.503 --> 00:32:14.044
People care about, they pay for the outcome, the product.

531
00:32:14.564 --> 00:32:17.466
I actually disagree about the process because...

532
00:32:18.366 --> 00:32:19.427
Matt Hughes, for example.

533
00:32:19.507 --> 00:32:19.987
Your editor.

534
00:32:20.127 --> 00:32:20.287
Yeah.

535
00:32:20.407 --> 00:32:20.568
Yeah.

536
00:32:20.988 --> 00:32:25.751
Watching him go down a rabbit hole and being there with him, and actually vice versa, him doing the same thing.

537
00:32:26.091 --> 00:32:29.493
We wrote these... Well, I mean, we were working on the research.

538
00:32:29.513 --> 00:32:35.697
I ended up sitting there for the day-long session of writing 11,000 words, and he had given me a bunch of notes.

539
00:32:35.717 --> 00:32:36.818
It was actually just...

540
00:32:36.838 --> 00:32:43.723
Even describing that process, I feel so happy, because it was like us being like, I can't believe how fuck these... Jesus Christ, they can't... Just like the...

541
00:32:44.769 --> 00:32:52.635
misanthropy of just the horrible, cynical people of asset managers like Blackstone, just learning about them and being like, it can't be this, and having a back and forth with him.

542
00:32:53.015 --> 00:32:56.177
That is fundamentally different because we were both learning together.

543
00:32:56.457 --> 00:33:00.540
And the learning process was as much about creating the output as the output itself.

544
00:33:00.821 --> 00:33:07.385
When you learn something, you're not creating the average, which really is what these things do, of the documents it could find.

545
00:33:07.826 --> 00:33:09.407
You're not getting particularly novel

546
00:33:10.807 --> 00:33:21.250
If I needed a generic slop output, sure, but I've used some of the higher-end LLM harness machines that the hedge funds use, and they all give the same shite.

547
00:33:21.450 --> 00:33:28.012
It's all the same generic reports, the same, oh, we noticed this analysis, things that you can find on any kind of AI slop out there.

548
00:33:28.172 --> 00:33:32.895
What you described to me there, what I heard anyway, is there's two points of value you're getting from your time with that.

549
00:33:33.275 --> 00:33:38.238
I mean, there's many more, but you said you're learning and then you're getting this book edited.

550
00:33:38.438 --> 00:33:38.678
Blog.

551
00:33:38.938 --> 00:33:39.158
Blog.

552
00:33:39.238 --> 00:33:42.720
You're getting a blog edited, which is the output, and you're getting learning.

553
00:33:42.740 --> 00:33:45.722
And you're also really getting connection and all these other things.

554
00:33:45.742 --> 00:33:49.164
But when people sort of think about the value of AI, of course they could use it to learn.

555
00:33:49.544 --> 00:33:54.407
But in the example I gave of like repeat my revenue number back to me or do this number, I just care about the output.

556
00:33:54.507 --> 00:33:55.347
I could use it to learn.

557
00:33:55.427 --> 00:33:56.188
I could say as far with me.

558
00:33:56.208 --> 00:33:57.828
What if the revenue number was wrong once?

559
00:33:58.149 --> 00:34:02.110
You should have defined, deterministic ways of knowing those numbers.

560
00:34:02.551 --> 00:34:03.751
You should not rely on them.

561
00:34:04.051 --> 00:34:10.154
Even with the terminal running BQL, which I trust, I will double-triple-triple-check everything just to be sure.

562
00:34:10.374 --> 00:34:14.396
Partly because also the process of learning for me, I don't want just a report I...

563
00:34:15.216 --> 00:34:26.779
go like that i want something that i fully understand and also understand the context around it i don't think that llms do that and i just don't see them getting better in a way that does that because it's

564
00:34:28.294 --> 00:34:29.434
It's just not what they do.

565
00:34:29.574 --> 00:34:34.236
And also there's the other problem of the more detailed the report, the more likely there are things to be wrong with it.

566
00:34:34.596 --> 00:34:38.337
If you are with Matt Hughes, for example, I can trust he's got it right.

567
00:34:38.617 --> 00:34:39.858
I can trust he understood.

568
00:34:40.098 --> 00:34:44.179
And I can trust that I can have a back and forth with him that will inform me if I've missed something.

569
00:34:44.539 --> 00:34:50.021
I can read the stuff that he's read and actually trust him because there's a big trust part as well.

570
00:34:50.341 --> 00:34:52.302
What is the basis of your trust in Matt?

571
00:34:52.402 --> 00:34:55.583
Could it be his historical performance?

572
00:34:56.083 --> 00:34:56.904
I mean, yes.

573
00:34:57.224 --> 00:34:57.524
Okay.

574
00:34:57.724 --> 00:35:00.386
And also the fact we've learned half of this stuff together.

575
00:35:00.446 --> 00:35:02.668
But tenure doesn't necessarily...

576
00:35:02.808 --> 00:35:05.209
There's probably people, you know, for 15 years who you also don't trust.

577
00:35:05.250 --> 00:35:05.470
Yes.

578
00:35:05.690 --> 00:35:09.733
So I think I was trying to figure out, like, what is the thing that's causing humans to trust another thing?

579
00:35:10.153 --> 00:35:13.956
And I guess it would be continual delivery of a commitment made of sorts.

580
00:35:14.616 --> 00:35:22.258
And so with Claude, for example, on simple tasks, as we've seen from this hallucination leaderboard, it continually delivers for people.

581
00:35:22.478 --> 00:35:25.178
And that's why we've seen the fastest- I mean, is that what that board says?

582
00:35:25.498 --> 00:35:27.699
Well, it's saying, like, is it getting it wrong?

583
00:35:27.899 --> 00:35:28.619
Is it hallucinating?

584
00:35:28.639 --> 00:35:29.499
On simple tasks.

585
00:35:29.579 --> 00:35:30.159
Simple tasks, yeah.

586
00:35:30.179 --> 00:35:30.919
How are those defined?

587
00:35:31.019 --> 00:35:31.699
I don't know.

588
00:35:31.719 --> 00:35:32.440
That's the thing, though.

589
00:35:32.640 --> 00:35:36.200
Because this is actually a very illustrative thing of the AI industry.

590
00:35:36.900 --> 00:35:39.581
They are the whataboutist masters.

591
00:35:39.621 --> 00:35:43.242
They have like, well, look, we've got this benchmark that says we're good at this.

592
00:35:43.262 --> 00:35:44.262
And look, the number's higher.

593
00:35:44.282 --> 00:35:44.422
Yeah.

594
00:35:44.462 --> 00:35:46.563
What's the number mean?

595
00:35:47.004 --> 00:35:47.544
What's that mean?

596
00:35:47.584 --> 00:35:49.385
And I'm not using this as a critique against you.

597
00:35:49.505 --> 00:35:53.568
It's when you can't give a direct answer, you give a side answer.

598
00:35:53.708 --> 00:35:58.331
When you as the LLM industry want to prove your worth, you can't just be like, just use the product.

599
00:35:58.391 --> 00:36:01.393
When the first iPhone came out, I got Penn State at the time.

600
00:36:01.894 --> 00:36:04.315
Oh, I felt like the apes at the beginning of 2001.

601
00:36:04.335 --> 00:36:05.276
I was like, oh, fuck.

602
00:36:05.902 --> 00:36:06.882
Visual voicemail.

603
00:36:07.143 --> 00:36:07.703
It was immediate.

604
00:36:07.723 --> 00:36:09.704
And I showed it to tech friends.

605
00:36:10.024 --> 00:36:11.824
I showed it to the most normal people in the world.

606
00:36:12.065 --> 00:36:13.485
And everyone was like, holy shit.

607
00:36:13.865 --> 00:36:14.706
They were on Razors.

608
00:36:14.726 --> 00:36:16.266
They were on Nokia 3210s.

609
00:36:16.606 --> 00:36:17.547
It was obvious the value.

610
00:36:17.807 --> 00:36:19.247
Amazon Web Services, same deal.

611
00:36:19.387 --> 00:36:20.288
It wasn't obvious, though.

612
00:36:20.488 --> 00:36:20.968
Yes, it was.

613
00:36:21.208 --> 00:36:22.449
I mean, I bought it.

614
00:36:22.569 --> 00:36:23.109
To you, it was.

615
00:36:23.289 --> 00:36:23.609
It was.

616
00:36:23.689 --> 00:36:28.011
And I also showed it to a bunch of people because I'm aware that I have bias when I just love gadgets.

617
00:36:28.251 --> 00:36:35.114
But I remember the famous Steve Barmer, who was the CEO of Microsoft, interview, where he was told about the iPhone.

618
00:36:35.754 --> 00:36:53.402
and he'd busts out laughing five hundred dollars fully subsidized with the plan i said that is the most expensive phone in the world and it doesn't appeal to business customers because it doesn't have a keyboard which makes it not a very good email machine you can get uh...

619
00:36:53.790 --> 00:36:56.814
a Motorola Q phone now for $99.

620
00:36:56.914 --> 00:36:58.576
It's a very capable machine.

621
00:36:58.977 --> 00:36:59.978
It'll do music.

622
00:37:00.038 --> 00:37:01.320
It'll do internet.

623
00:37:01.360 --> 00:37:02.361
It'll do email.

624
00:37:02.441 --> 00:37:04.083
It'll do instant messaging.

625
00:37:04.484 --> 00:37:08.369
So I kind of look at that and I say, well, I like our strategy.

626
00:37:08.449 --> 00:37:09.250
I like it a lot.

627
00:37:10.477 --> 00:37:11.617
He burst out laughing, mocking it.

628
00:37:11.877 --> 00:37:13.398
Because it was so disruptive.

629
00:37:13.518 --> 00:37:14.718
It was way more expensive.

630
00:37:15.378 --> 00:37:16.338
And it was way different.

631
00:37:16.378 --> 00:37:16.959
No keyboard.

632
00:37:17.219 --> 00:37:21.820
Well, phones used to be insanely expensive and the carriers would cover them, but you had to sign a long contract.

633
00:37:22.040 --> 00:37:23.340
You were still spending 500 bucks.

634
00:37:23.460 --> 00:37:30.142
But the thing I'm getting at is you didn't have to explain to someone what, perhaps you'd have to get past the cost part, but you could just be like, look how good this is.

635
00:37:30.602 --> 00:37:35.186
And then once the app was the iPhone 3G with the App Store, people were like, oh shit, this could actually change things.

636
00:37:35.446 --> 00:37:39.429
Mobile web, even though it was a monstrosity, it was so bad at first.

637
00:37:39.889 --> 00:37:41.751
Even then you could get your emails and you could just look at them.

638
00:37:42.071 --> 00:37:44.773
But it's BlackBerrys were also expensive and were still actually kind of cool.

639
00:37:45.233 --> 00:37:47.615
But the way they worked was not like consumer software.

640
00:37:47.635 --> 00:37:48.956
They didn't have the classic GUI.

641
00:37:49.336 --> 00:37:50.117
iPhones felt like that.

642
00:37:50.177 --> 00:37:54.200
It felt like a cell phone designed even, like a computer.

643
00:37:54.500 --> 00:37:55.261
It was obvious.

644
00:37:55.361 --> 00:37:56.482
It was obvious in the beginning.

645
00:37:56.682 --> 00:37:57.403
Everyone, I was, I was,

646
00:37:57.983 --> 00:38:01.947
day and a go in the center of Pennsylvania at the time and everyone I showed it to was like, wow, this is incredible.

647
00:38:02.348 --> 00:38:04.310
That, to me, is the obvious thing.

648
00:38:04.570 --> 00:38:08.374
With AI, to this day, when you're like, okay, why is it so amazing?

649
00:38:08.794 --> 00:38:09.615
People still dither.

650
00:38:09.836 --> 00:38:13.579
People are still like, yeah, you can't run a business fully with it without this...

651
00:38:14.540 --> 00:38:17.562
weird system of pulleys and levers and such.

652
00:38:17.702 --> 00:38:26.968
But how come then, when you look at the stats around ChatGBT's growth, 100 million active users in just the first 60 days after launching?

653
00:38:27.048 --> 00:38:33.713
For comparison, TikTok took nine months, Instagram took 2.5 years, and the internet itself, the World Wide Web, took roughly seven years to reach that scale.

654
00:38:34.093 --> 00:38:42.939
Over 60% of the US adults are integrated into AI tools in their daily and regular routines within three years of the launch, reaching 40% of the population population.

655
00:38:43.719 --> 00:38:47.921
And that same milestone took the internet five years and personal computers nearly 12.

656
00:38:48.441 --> 00:38:48.781
Okay.

657
00:38:49.061 --> 00:38:51.402
So I think this is the part that's giving me dissonance.

658
00:38:51.422 --> 00:38:55.103
It's like when I showed my fiancé ChatGVT, it didn't really work.

659
00:38:55.884 --> 00:39:11.690
But as a sole entrepreneur who English isn't her first language, who has to write lots of text and lots of copy and generate lots of images and was paying a graphic designer to help her make certain images that she couldn't make herself because she doesn't have the skills, she would describe it as being...

660
00:39:12.650 --> 00:39:38.343
transformative for her business what i'm hearing from you is that it's not transformative and there's no value in it for people but she if she was sat here now she could transformative it would she pay the per million token rate would she pay the actual rate because that's the thing if this was sold at its honest cost yeah i would actually if and people were reacting like that and they were paying two three four dollars every time they did something and they were genuinely happy that might be an argument what is the what would be the honest cost if they weren't

661
00:39:38.503 --> 00:39:41.545
The actual per million token cost, the actual API cost.

662
00:39:41.565 --> 00:39:42.105
They should charge.

663
00:39:42.145 --> 00:39:44.606
Do you know how much that is relative to what they charge?

664
00:39:44.626 --> 00:39:44.826
Oh, God.

665
00:39:44.866 --> 00:39:45.827
It depends on the model.

666
00:39:45.967 --> 00:39:51.409
But there's actually kind of a point I want to make about the thing you said with the internet earlier.

667
00:39:51.689 --> 00:39:55.471
So when I first got on the internet, a 33.4 kilobits a second modem.

668
00:39:56.332 --> 00:39:58.793
Even back then, I was like, fuck, if this was faster...

669
00:39:59.593 --> 00:40:02.815
And that was like immediate, just like if this was faster, because it was slow.

670
00:40:02.855 --> 00:40:07.879
You'd go on like Happy Puppy or something, download, take all bloody day waiting for shareware to download.

671
00:40:08.219 --> 00:40:10.401
Immediately, you were like, if I could do this faster, it would be better.

672
00:40:10.461 --> 00:40:13.863
And even back then, I'm like, man, you could probably do video camera stuff with this.

673
00:40:14.303 --> 00:40:15.464
Stuff that eventually happened.

674
00:40:15.484 --> 00:40:22.229
And actually, there's this guy called Jim Covello from Goldman Sachs in a report he did in 2024 that was Gen.AI, too much spend for not enough return.

675
00:40:22.649 --> 00:40:23.450
Paraphrasing there.

676
00:40:23.470 --> 00:40:25.672
And he made the point that in the run up to the iPhone, right?

677
00:40:26.492 --> 00:40:36.300
There was thousands of presentations that when GSM radios get smaller, when Bluetooth radios get smaller, when Wi-Fi radios get smaller, it is inevitable that we will get something like this.

678
00:40:36.661 --> 00:40:39.943
And then he said that there is no such path for AI.

679
00:40:39.963 --> 00:40:42.626
There was no roadmap to AI becoming...

680
00:40:43.166 --> 00:40:44.227
This thing that they promised.

681
00:40:44.287 --> 00:40:49.570
And I must be clear, if these companies had gone out there and are like, yeah, this is interesting cloud software.

682
00:40:49.590 --> 00:40:50.170
It's generative.

683
00:40:50.411 --> 00:40:51.571
It's really expensive.

684
00:40:52.012 --> 00:40:56.274
We're not sure if we can fully not trust it, not in the oh, I'm scared way.

685
00:40:56.294 --> 00:41:00.077
I mean, just like we're not sure that this is going to be a disruptive...

686
00:41:00.517 --> 00:41:01.238
world-changing thing.

687
00:41:01.358 --> 00:41:03.039
It has potential, but we're going to go slow.

688
00:41:03.259 --> 00:41:04.340
It's really expensive.

689
00:41:04.380 --> 00:41:05.601
This is an R&D effort.

690
00:41:05.862 --> 00:41:13.088
We're not going to expose consumers to it and actually be like, I don't know, language models and no generative AI stuff.

691
00:41:13.208 --> 00:41:15.410
Just being, not even calling it because it isn't AI.

692
00:41:15.610 --> 00:41:16.391
It's not autonomous.

693
00:41:16.431 --> 00:41:17.152
It's not smart.

694
00:41:17.212 --> 00:41:17.652
It's not smart.

695
00:41:17.772 --> 00:41:19.874
I actually might respect it, but this is not.

696
00:41:19.934 --> 00:41:24.839
They've gone out there since 2023 and said, 2022, this is the best thing since sliced bread.

697
00:41:24.859 --> 00:41:26.080
This is changing everything.

698
00:41:26.300 --> 00:41:27.561
This is going to do all your work.

699
00:41:27.581 --> 00:41:29.203
This is going to take your job.

700
00:41:29.503 --> 00:41:31.625
You're going to talk to Bing and it's going to take you to leave your wife.

701
00:41:31.705 --> 00:41:33.006
All of these crazy things.

702
00:41:33.747 --> 00:41:40.550
And what's funny is when the writer, Kevin Roos, I think it was, he was speaking to Kevin Scott, the CTO of Microsoft, about it.

703
00:41:40.570 --> 00:41:43.472
And Kevin Scott goes, you know, I'm just glad we're having this conversation.

704
00:41:43.532 --> 00:41:45.693
Instead of being like, settle down, Beavis.

705
00:41:46.193 --> 00:41:47.073
It's a website.

706
00:41:47.113 --> 00:41:48.374
The website told you something.

707
00:41:48.434 --> 00:41:49.214
It's just LLMs.

708
00:41:49.535 --> 00:41:50.175
They talked it up.

709
00:41:50.335 --> 00:41:55.557
And that's because everyone is talking about what they wish this was rather than talking about what it can actually do.

710
00:41:55.958 --> 00:41:58.759
This makes it scary to people, deliberately so.

711
00:41:59.399 --> 00:42:01.000
It makes it environmentally destructive.

712
00:42:01.140 --> 00:42:02.961
Look at the gas turbines poisoning blank neighborhoods.

713
00:42:03.321 --> 00:42:04.382
I think it's in Louisiana.

714
00:42:04.442 --> 00:42:06.083
It's one of Musk's data centers.

715
00:42:06.523 --> 00:42:08.204
Look at the incredible energy draws.

716
00:42:08.304 --> 00:42:09.605
It is raising power bills.

717
00:42:10.185 --> 00:42:15.648
And also, it is creating inflation across all consumer electronics because of the massive ramp.

718
00:42:15.668 --> 00:42:16.589
Do you know what's interesting?

719
00:42:16.729 --> 00:42:20.771
I almost feel like so much of what you're saying is true.

720
00:42:21.392 --> 00:42:27.315
And also, it can be true that this technology is going to profoundly change the world.

721
00:42:28.247 --> 00:42:31.369
And I think back to the early days of...

722
00:42:31.449 --> 00:42:35.332
The internet is maybe the closest analogy we have of... You know, in the dot-com bubble.

723
00:42:35.472 --> 00:42:36.753
You know, you wrote this great essay.

724
00:42:36.793 --> 00:42:37.254
Yes, yes.

725
00:42:37.774 --> 00:42:38.675
Which I found really funny.

726
00:42:39.736 --> 00:42:40.216
Especially the name.

727
00:42:40.676 --> 00:42:41.997
The Rot Economy.

728
00:42:42.117 --> 00:42:43.919
And you talked about the rot-com bubble.

729
00:42:44.279 --> 00:42:44.539
Yes.

730
00:42:44.900 --> 00:42:48.362
Talking about how AI is of less value than people think.

731
00:42:48.722 --> 00:42:48.863
Mm-hmm.

732
00:42:49.883 --> 00:42:57.069
And in the sort of dot-com bubble, what you saw is huge hype, people overselling the capabilities of their websites and what they were building.

733
00:42:57.509 --> 00:43:02.974
But in the wake of the dot-com bubble, yes, 90% of stuff went to zero.

734
00:43:03.655 --> 00:43:06.677
But you had generational companies born that changed the world.

735
00:43:06.697 --> 00:43:07.218
Right.

736
00:43:07.238 --> 00:43:07.538
Right.

737
00:43:07.898 --> 00:43:11.064
And so I do, I kind of, and that's what bubbles do, right?

738
00:43:11.585 --> 00:43:13.509
Huge hype, overinvestment.

739
00:43:13.669 --> 00:43:15.172
Investors get crazy delusional.

740
00:43:15.192 --> 00:43:16.635
They think everything's going to change.

741
00:43:17.296 --> 00:43:19.741
At the same time, you do have skeptics.

742
00:43:20.596 --> 00:43:22.657
In these moments, the dot-com bubble had...

743
00:43:22.697 --> 00:43:24.597
I mean, the internet itself had the biggest sceptics.

744
00:43:24.637 --> 00:43:35.481
In 1998, Nobel Prize-winning economist Paul Krugman said, By 2005 or so, it will become clear that the internet's impact on the economy has been no greater than the fax machine.

745
00:43:36.141 --> 00:43:42.984
In 1995, astrophysicist Clifford Stuhl, famously, I wrote about this in my book, wrote famously in Newsweek...

746
00:43:43.704 --> 00:43:46.107
Do our computer pundits lack all common sense?

747
00:43:46.287 --> 00:43:49.891
The truth is no online database will replace your daily newspaper.

748
00:43:50.031 --> 00:43:52.834
No CD-ROM can take the place of a competent teacher.

749
00:43:53.275 --> 00:43:57.079
Commerce and businesses will shift from offices and malls to networks and modems?

750
00:43:57.499 --> 00:43:58.140
Baloney.

751
00:43:58.280 --> 00:44:04.066
So how come my local mall does a roaring business and the cyber mall gets zero business?

752
00:44:04.727 --> 00:44:08.391
And then I'll give you one more from Krugerman, who was the award-winning economist.

753
00:44:09.032 --> 00:44:11.555
He said, the growth of the internet will slow drastically.

754
00:44:11.915 --> 00:44:14.759
As it becomes apparent, most people have nothing to say to each other.

755
00:44:15.800 --> 00:44:20.285
That may actually be the worst one of those predictions.

756
00:44:20.605 --> 00:44:22.607
Hang around any bar in middle America.

757
00:44:22.627 --> 00:44:23.969
Honestly, the best conversation.

758
00:44:23.969 --> 00:44:25.410
But it's just all the same thing.

759
00:44:25.730 --> 00:44:34.418
I actually, so Clifford Stoll actually, his piece was interesting because there were some boner points in it, but he made points about how like an overwhelming amount of bad information out there is bad for society.

760
00:44:34.438 --> 00:44:35.158
He's completely right.

761
00:44:35.599 --> 00:44:40.283
Saying how online education would not be a great replacement for regular education.

762
00:44:40.303 --> 00:44:41.183
I think we've seen that.

763
00:44:41.323 --> 00:44:41.584
But...

764
00:44:41.964 --> 00:44:45.786
There is an economic difference that's vastly, it's just completely different.

765
00:44:45.966 --> 00:44:47.888
So, dot-com bubble was actually two bubbles.

766
00:44:48.348 --> 00:44:51.210
There was the website bubble, which was just trash on trash on trash.

767
00:44:51.410 --> 00:44:53.271
It was just like, I think, what was it?

768
00:44:53.391 --> 00:44:57.934
Excite at Home bought an eager eating card company for like a billion dollars.

769
00:44:57.954 --> 00:44:59.114
It was insane crap happening.

770
00:44:59.535 --> 00:45:01.096
That was so small.

771
00:45:01.656 --> 00:45:04.398
The big thing that people are thinking about is the dark fiber.

772
00:45:04.518 --> 00:45:05.138
Dark fiber, yeah.

773
00:45:05.738 --> 00:45:10.240
Dark fiber was all of the wires that put in the ground thinking we're going to have all this demand for internet.

774
00:45:10.580 --> 00:45:20.883
And it turned out that demand for internet, I think the analyst estimate was it was doubling every 90 days when it was doing that every six to 12 months, maybe longer.

775
00:45:21.303 --> 00:45:29.845
And just thus there was a massive overbuild of fiber optic cable and indeed the transmission stations and such to simplifying to bring that to people's houses.

776
00:45:30.166 --> 00:45:33.927
And there was the assumption that, well, that would all get lit up and people would want it immediately.

777
00:45:34.107 --> 00:45:34.747
Didn't really happen.

778
00:45:35.287 --> 00:45:39.652
Now, the post.com bubble thing people say is, well, but after that, there was demand from the internet.

779
00:45:40.470 --> 00:45:41.350
That's the thing, though.

780
00:45:41.610 --> 00:45:44.391
That's very different to demand for generative AI.

781
00:45:44.851 --> 00:45:48.592
Right now, the demand we have for generative AI is predominantly subsidized.

782
00:45:48.692 --> 00:45:50.212
Just let's start there.

783
00:45:50.392 --> 00:45:54.293
Predominantly subsidized, and most people experience it and yet are not paying the real cost.

784
00:45:54.533 --> 00:45:54.853
I agree.

785
00:45:55.153 --> 00:46:00.114
On top of that, we already have all of the possible marketing in the world.

786
00:46:00.174 --> 00:46:04.615
We have the largest, most disingenuous marketing campaign in the history of man.

787
00:46:05.664 --> 00:46:06.585
pushing this up the hill.

788
00:46:06.765 --> 00:46:10.727
We have the apex predator of cloud software, Microsoft.

789
00:46:11.128 --> 00:46:15.350
They can only get single digit billions from selling AI software.

790
00:46:15.591 --> 00:46:19.593
And Christ almighty, outside of OpenAI and Anthropic, we barely get $22 billion.

791
00:46:19.994 --> 00:46:22.895
And the thing is, $22 billion is a large amount to you and me.

792
00:46:23.296 --> 00:46:31.281
It's not a large amount of money when you spent a trillion plus dollars, when you have Anthropic and OpenAI with $1.1 trillion worth of cloud commitments.

793
00:46:31.541 --> 00:46:32.182
And on top of that,

794
00:46:32.842 --> 00:46:35.204
How does this turn into a post.com bubble thing?

795
00:46:35.584 --> 00:46:43.129
A data center built today is going to be as expensive to run in 2050 as it is today, unless there's some breakthrough in electricity.

796
00:46:43.270 --> 00:46:45.351
But again, that's not happening with AI.

797
00:46:45.551 --> 00:46:49.074
AI is not doing that unless there's some breakthrough in GPU technology.

798
00:46:49.174 --> 00:46:52.096
But we already have Broadcom, NVIDIA, Etched.

799
00:46:52.356 --> 00:46:56.679
We have every major chip company, ARM, trying to do something about this.

800
00:46:57.160 --> 00:47:01.803
And no one seems to magically be able to make this profitable or indeed even less costly.

801
00:47:01.823 --> 00:47:01.983
Right.

802
00:47:02.163 --> 00:47:08.928
Even NVIDIA with Vera Rubin, their more expensive new GPU system, even then they're like, yeah, 10x more efficient.

803
00:47:09.048 --> 00:47:10.770
It's more dollars per megawatt.

804
00:47:10.950 --> 00:47:11.931
They're all coy about it.

805
00:47:11.991 --> 00:47:16.494
They don't just say, yeah, we worked with OpenAI and Anthropic and we found it reduced their cost by 50%.

806
00:47:16.834 --> 00:47:18.175
Easiest thing in the world if it was true.

807
00:47:18.656 --> 00:47:20.417
And that's because it's not happening.

808
00:47:20.877 --> 00:47:22.399
And this isn't a case where...

809
00:47:22.579 --> 00:47:25.001
So you're saying there's not going to be the demand for...

810
00:47:25.681 --> 00:47:29.146
Let's say there's different types of AI, generative AI.

811
00:47:29.266 --> 00:47:30.968
Yeah, and actually that's a good point to make.

812
00:47:31.189 --> 00:47:35.394
The reason they use the term artificial intelligence is so everyone would lump everything into it.

813
00:47:35.935 --> 00:47:38.899
They would lump protein folding, nothing to do with LLMs.

814
00:47:39.079 --> 00:47:40.361
Robotics, not LLMs.

815
00:47:41.142 --> 00:47:44.865
Autonomous weapons, even horrible as they are, not LLMs because you couldn't trust them.

816
00:47:45.565 --> 00:47:52.771
But they've mushed everything into AI so that when you say, well, AI can't, they'll go, um, um, sir, you forgot to give us homework.

817
00:47:52.791 --> 00:47:57.254
And also AI, it's working on curing cancer when it's just like, no, that's not LLMs.

818
00:47:57.434 --> 00:47:58.295
Stop giving them credits.

819
00:47:58.435 --> 00:48:01.316
The similarity, though, is they all need GPUs, all these AIs.

820
00:48:01.336 --> 00:48:02.417
No, and that's the funny thing.

821
00:48:02.677 --> 00:48:07.059
All those data centers that we're building, all of them are for just generative AI.

822
00:48:07.519 --> 00:48:09.980
They're not for all of the other stuff.

823
00:48:10.280 --> 00:48:11.521
They're not for the cool shit.

824
00:48:11.761 --> 00:48:13.581
AI has been around for a long time.

825
00:48:14.142 --> 00:48:19.584
Google, a lot of the good stuff that comes out of Google from the search side is AI, but pre-generative.

826
00:48:19.724 --> 00:48:21.885
How would you run the type of...

827
00:48:22.125 --> 00:48:26.751
AI that sits in a robot, let's say one of the Optimus robots, if you didn't have a GPU.

828
00:48:27.112 --> 00:48:29.755
So Matic, Matic has this cleaning robot, for example.

829
00:48:30.116 --> 00:48:31.998
That thing has not got a little GPU in it.

830
00:48:32.767 --> 00:48:42.670
What it has, and may indeed have used some GPUs, but now a year as many as they need for generative AI to run the data, feed-training data into it so it's able to clean a house.

831
00:48:42.710 --> 00:48:49.591
But when the little bugger's going around cleaning my floor, turdsly I call it, goes around mopping my floor, it's not like burning money the whole time.

832
00:48:49.931 --> 00:48:59.674
But when it comes to these massive amount of data centers, Sightline Climate said in February, there's 190 gigawatts of data centers under planning, don't know about under construction.

833
00:49:00.394 --> 00:49:06.957
That works out if about 12 million megawatts, well, like $1.6 trillion to $3 trillion a year in annual demand you'd need for that.

834
00:49:07.417 --> 00:49:10.339
We don't even have $130 billion worth of annual demand.

835
00:49:10.359 --> 00:49:11.560
And people say, well, it will grow.

836
00:49:12.000 --> 00:49:12.360
How?

837
00:49:12.600 --> 00:49:22.065
When most of the demand is coming from Amazon feeding money to open AI or Anthropic, Microsoft feeding money to open AI and Anthropic, Google feeding money to open AI and Anthropic.

838
00:49:22.245 --> 00:49:26.427
Well, Google hasn't fed it to open AI yet, but they're a pretty big customer, billions of dollars.

839
00:49:26.967 --> 00:49:28.088
The con side is that,

840
00:49:28.848 --> 00:49:32.851
We are building these effigies to capitalism, these giant GPU data centers.

841
00:49:33.231 --> 00:49:35.933
And people are being told, well, it's for AI.

842
00:49:36.453 --> 00:49:38.714
You know, the thing that's done all this other stuff that's unrelated.

843
00:49:39.175 --> 00:49:43.517
Or the worst thing I've seen, it's like, oh, you like online banking, where you do like data centers.

844
00:49:43.758 --> 00:49:54.625
There's a big difference between a data center for regular non-GPU compute for standing up a server, a content delivery system like Akamai or something that brings the website to you, or how Meta runs Facebook.

845
00:49:55.305 --> 00:49:56.306
That is not the same.

846
00:49:56.346 --> 00:50:02.030
It takes way less power, mostly CPU-driven, compared to these giant GPU data centers that are for one thing.

847
00:50:02.790 --> 00:50:03.330
One thing, aren't they?

848
00:50:03.370 --> 00:50:06.473
But I was doing the research and looking at some of these notes here.

849
00:50:06.853 --> 00:50:16.439
It does say that for tougher types of AI systems designed to solve concrete physics, biology, and spatial problems, they require some of the most intense data center infrastructure on the planet.

850
00:50:17.120 --> 00:50:24.485
AI systems like DeepMind's AlphaFold, the protein folding company, used for genomic sequencing and climate forecasting, etc.,

851
00:50:25.025 --> 00:50:26.966
run on high-performance computing clusters.

852
00:50:27.026 --> 00:50:30.528
These require immense precision and continuous heavy computing data centers.

853
00:50:31.208 --> 00:50:38.612
Training the brains for self-driving cars requires billions of miles of simulated physics environments.

854
00:50:38.712 --> 00:50:40.373
The AI isn't generating text.

855
00:50:40.413 --> 00:50:44.595
It's learning to navigate 3D spaces and gravity and relies on data centers.

856
00:50:44.675 --> 00:50:44.955
Right.

857
00:50:45.175 --> 00:50:48.317
And the thing is, those data centers, they might have GPUs in them.

858
00:50:48.537 --> 00:50:54.200
We had GPUs used for this HPC, the high-performance computing, before generative AI.

859
00:50:54.440 --> 00:50:56.562
And yeah, that's how AI has been trained before.

860
00:50:56.602 --> 00:50:57.844
That's how Tesla did.

861
00:50:57.904 --> 00:51:02.889
I believe they've had their own data centers when it comes to training the autopilot system, for better or for worse.

862
00:51:03.209 --> 00:51:04.110
That's how we've done it before.

863
00:51:04.371 --> 00:51:07.354
Again, that is not why we're building these data centers.

864
00:51:07.394 --> 00:51:14.241
These data centers are being built to sell to generative AI companies to either train systems or run inference.

865
00:51:14.962 --> 00:51:22.286
These things are being built in this brainless way, where it's just, well, actually, maybe this is a good way of illustrating the con.

866
00:51:22.967 --> 00:51:37.115
Because everyone saw Google, Microsoft, Amazon, and Meta give NVIDIA over, call it 800-something billion dollars, because everyone saw that, they went, well, they wouldn't do that for no reason.

867
00:51:37.155 --> 00:51:41.278
They went, we've got to build more of these things, because there must be all this demand, even though the demand...

868
00:51:42.256 --> 00:51:47.721
70% or more of all that demand comes from these two companies who are funded by these three companies.

869
00:51:48.201 --> 00:51:49.162
And that's the funny thing.

870
00:51:50.143 --> 00:51:57.108
The reason that they don't want to break out their AI revenues is because it will become alarmingly obvious that this was the case.

871
00:51:57.329 --> 00:52:02.413
It turns out that the only real big customers, because it's not like they're building a few data centers.

872
00:52:02.553 --> 00:52:05.916
They're building trillion plus revenue potential.

873
00:52:06.036 --> 00:52:06.877
They believe they'll get.

874
00:52:07.337 --> 00:52:07.958
Speculative.

875
00:52:07.978 --> 00:52:08.658
Entirely speculative.

876
00:52:08.678 --> 00:52:09.159
Speculative.

877
00:52:09.779 --> 00:52:14.765
They're building it because they saw the biggest companies in the world buy a bunch of GPS and they said, I want in on that.

878
00:52:14.985 --> 00:52:16.707
They must have diverse customers, right?

879
00:52:16.967 --> 00:52:20.872
They wouldn't just have two unprofitable fail sons that they're propping up with.

880
00:52:21.593 --> 00:52:23.154
Christ, they've raised $217 billion just in 2026.

881
00:52:26.893 --> 00:52:32.956
So we know that some of the biggest companies in the world are using AI, generative AI, to write a lot of their code.

882
00:52:33.656 --> 00:52:36.418
That is a great productivity gain for those companies, right?

883
00:52:36.818 --> 00:52:41.780
I mean, have you used Google or Facebook or Instagram or GitHub recently?

884
00:52:41.820 --> 00:52:42.521
Because they are...

885
00:52:43.141 --> 00:52:44.041
catastrophically worse.

886
00:52:44.141 --> 00:52:48.603
Amazon Web Services went down multiple times because of their AI coding tool.

887
00:52:48.823 --> 00:52:49.643
How is Google worse?

888
00:52:49.883 --> 00:52:53.144
Well, I'll tell the story of a real arsehole, a guy called Prabhagar Raghavan.

889
00:52:53.304 --> 00:52:55.284
Previously, one of the heads of ads at Google.

890
00:52:55.404 --> 00:53:00.706
In 2019, Google called something called a code yellow, which is when they said, we've got a problem.

891
00:53:00.726 --> 00:53:07.268
And it was material weakness in query numbers, which means the amount of times that people were searching on Google search.

892
00:53:07.768 --> 00:53:11.509
Guy called Ben Gomes, internal at Google, then the head of Google search, says, okay,

893
00:53:11.709 --> 00:53:22.637
wait a minute, to increase this number of using Google more, we're going to have to, I mean, what you're suggesting would mean we give worse answers, because if someone got the answer quickly, that would reduce the amount of queries, right?

894
00:53:23.198 --> 00:53:28.421
And people at Google, Shashi Thakur was another engineer who was saying, yeah, can we please tell Sundar this?

895
00:53:28.702 --> 00:53:30.143
Because this doesn't seem good.

896
00:53:30.163 --> 00:53:32.865
We can't just increase the amount of queries.

897
00:53:32.905 --> 00:53:35.667
That would just mean that people would have to search more, which would make the product worse.

898
00:53:36.287 --> 00:53:37.848
But it would make them more money, are you saying?

899
00:53:37.908 --> 00:53:38.048
Yes.

900
00:53:38.188 --> 00:53:40.069
Because you'd show them more ads.

901
00:53:40.270 --> 00:53:42.771
So if you were spending more time on Google because Google's work...

902
00:53:42.871 --> 00:53:44.652
But is this linked to AI doing code?

903
00:53:44.772 --> 00:53:45.353
Oh, I'll get there.

904
00:53:46.093 --> 00:53:47.714
So this is...

905
00:53:48.114 --> 00:53:55.099
The problem is that this guy called Prabhupar Raghavan, who was the head of ads at the time, was pushing, pushing and saying, no, we need to make more queries happen.

906
00:53:55.259 --> 00:53:55.999
Got to make it happen.

907
00:53:56.019 --> 00:53:59.982
Nick Fox, who was there as well, I believe he's actually taking over Google search, said, got to make them go up.

908
00:54:00.002 --> 00:54:00.982
This is our new reality.

909
00:54:01.003 --> 00:54:01.623
Yeah.

910
00:54:02.063 --> 00:54:05.585
Sometime in early 2020, Prabhagar Raghavan takes over Google search.

911
00:54:05.946 --> 00:54:19.735
From then, and this is what I believe, can't prove it, if you go and look around the various SEO sites such as a journal and the various forums, Google stripped back a lot of the suppression of spammy sites so that people would be on Google more.

912
00:54:20.415 --> 00:54:26.757
And then over the course of time, Google wanted to create more queries, and Google search became much worse.

913
00:54:26.777 --> 00:54:29.918
It's why people always do, like, plus Reddit, or from Reddit, or what have you.

914
00:54:30.178 --> 00:54:32.939
It's because the actual underlying search results of Google had got worse.

915
00:54:33.399 --> 00:54:34.699
And then generative AI came along.

916
00:54:35.159 --> 00:54:36.660
And Prabhbagar, wouldn't you know, it gets...

917
00:54:37.280 --> 00:54:38.341
put to run part of Gemini.

918
00:54:38.701 --> 00:54:42.403
And Google also was having trouble getting people back on Google.

919
00:54:42.663 --> 00:54:43.864
And what did they think they'd do?

920
00:54:44.024 --> 00:54:46.005
Well, shit, everyone's talking about this AI thing.

921
00:54:46.605 --> 00:54:47.826
We'll just put it right at the top.

922
00:54:47.846 --> 00:54:49.266
So people have to stay at Google.

923
00:54:49.707 --> 00:55:01.413
And actually they'll use it more because instead of searching websites and doing that annoying thing where they click away from Google, they'll just only use Google instead of generating answers, by which I mean, giving you search results you click through.

924
00:55:01.733 --> 00:55:02.874
Now Google is the answer.

925
00:55:03.134 --> 00:55:03.634
Is it right?

926
00:55:03.714 --> 00:55:04.054
God, no.

927
00:55:04.094 --> 00:55:05.235
It might tell you to eat rocks.

928
00:55:05.655 --> 00:55:28.852
might eat poisonous mushrooms maybe i'll give you a little few links you could click through but the ideal situation was that ai was the ultimate form of google's evil which was but i'm saying here i'm saying here but that's not the fact that coders could code on google that's made google worse that's human decisions have made it worse yes and then there's the instability of google's platform which is actually i should have probably led with that a problem across the whole tech industry

929
00:55:29.397 --> 00:55:32.419
Okay, so you're saying Google is going down more.

930
00:55:32.519 --> 00:55:33.920
Yes, Google is less stable.

931
00:55:34.220 --> 00:55:37.622
Google Docs is a bug fest right now and has been for a while.

932
00:55:37.662 --> 00:55:38.703
Google Sheets, same deal.

933
00:55:39.123 --> 00:55:40.444
And the thing is, you're right.

934
00:55:40.664 --> 00:55:41.564
I'm being a little unfair.

935
00:55:41.724 --> 00:55:42.445
This is everyone.

936
00:55:42.765 --> 00:55:43.826
It's the same with Microsoft.

937
00:55:43.866 --> 00:55:44.806
It's the same with Amazon.

938
00:55:44.866 --> 00:55:45.727
It's the same across the world.

939
00:55:45.747 --> 00:55:47.588
How do we quantify that outside of anecdotes?

940
00:55:47.968 --> 00:55:49.449
Like, is there a way to... You're right.

941
00:55:49.589 --> 00:55:51.870
I mean, GitHub downtime is the best example.

942
00:55:52.031 --> 00:55:57.094
Amazon Web Services went down, I think, two or three times this year because of AI tools.

943
00:55:57.294 --> 00:55:57.734
Wow.

944
00:55:57.874 --> 00:55:58.794
And honestly, you're right.

945
00:55:59.055 --> 00:56:01.416
It's kind of hard to quantify outside of anecdotes.

946
00:56:01.616 --> 00:56:05.537
But I challenge anyone listening to this, go and use the website these days and tell me how well it works.

947
00:56:06.018 --> 00:56:06.918
Tell me how buggy it is.

948
00:56:07.218 --> 00:56:14.802
Tell me how many problems, even with my iPhone, the supposed best UX in town, even the iPhone is a flipping mess these days.

949
00:56:15.382 --> 00:56:17.423
Okay, so the research says...

950
00:56:18.323 --> 00:56:19.943
The short answer is yes.

951
00:56:20.484 --> 00:56:29.586
Tech downtime and software outages have demonstrably increased over the last few years, and industry data points directly to the explosion of AI-assisted coding as a primary culprit.

952
00:56:30.086 --> 00:56:32.866
The problem is hitting the tech industry from two entirely different directions.

953
00:56:33.267 --> 00:56:39.948
The code itself is getting buggier, and the sheer volume of AI activity is literally crashing the underlying infrastructure.

954
00:56:40.468 --> 00:56:40.848
Interesting.

955
00:56:41.068 --> 00:56:44.789
Yeah, that's because GitHub, people are just writing a bunch of code, pushing it,

956
00:56:48.310 --> 00:56:48.970
That's interesting.

957
00:56:49.130 --> 00:56:55.333
Yeah, it's a real mess as well because open sources have this problem as well because it's well-meaning people.

958
00:56:55.354 --> 00:56:57.274
They're like, oh, I learned a bit of code with an LLM.

959
00:56:57.294 --> 00:56:58.695
I'm going to go out and do some stuff.

960
00:56:58.715 --> 00:56:59.816
I'm going to make this project better.

961
00:56:59.836 --> 00:57:02.457
And these people barely understand what they're shipping.

962
00:57:02.917 --> 00:57:07.259
Or maybe they understand a bit of code and they say, oh, Dunning-Kruger, this motherfucker.

963
00:57:07.559 --> 00:57:09.120
I'm just like, I can understand some of this.

964
00:57:09.140 --> 00:57:11.241
And now the code's all written and just push it right now.

965
00:57:11.541 --> 00:57:13.502
So GitHub is flooded with AI code.

966
00:57:13.542 --> 00:57:16.404
This sounds like it's making humans complacent.

967
00:57:16.944 --> 00:57:17.204
It is.

968
00:57:17.425 --> 00:57:22.070
Because we're going, okay, look, I'll let it write the code for the last 100 lines.

969
00:57:22.391 --> 00:57:23.352
And it was broadly right.

970
00:57:23.432 --> 00:57:25.515
So the next 100 lines, I won't check them as much.

971
00:57:25.535 --> 00:57:25.915
Yeah, yeah, yeah.

972
00:57:26.296 --> 00:57:32.924
And that's human nature, is to get sort of to take shortcuts, to spend less energy on an activity if you can.

973
00:57:33.204 --> 00:57:37.565
Right, but the AI's still making the mistake and we're still making all the promises of AI.

974
00:57:37.585 --> 00:57:38.106
That's the thing.

975
00:57:38.786 --> 00:57:42.307
This thing is meant to be this autonomous... You say it can't be perfect.

976
00:57:42.327 --> 00:57:43.107
I don't know.

977
00:57:43.687 --> 00:57:49.889
Based on what Sam Altman has been saying for the last few years, Clammy Sammy's been promising the world, saying this will replace software engineers.

978
00:57:50.289 --> 00:57:58.012
Dario Amadei, Wario himself, has been saying, oh yeah, 50% of white-collar labor is going to go away in the next few years.

979
00:57:58.292 --> 00:57:59.833
These people are promising the world.

980
00:57:59.933 --> 00:58:00.953
Again, if they were...

981
00:58:01.393 --> 00:58:26.216
saying it would be smaller and they were like yeah it does have issues and we must be none of this oh what if it wakes up and it's super powerful just like yeah it's probabilistic it's going to make mistakes and if you don't know what you're doing you don't really know what you're looking at you're going to miss those mistakes and it's going to get multiplicatively worse as you go when you don't know what you're doing so yeah human nature is part of it but so is the marketing so are the promises

982
00:58:27.320 --> 00:58:34.147
One of the smartest things a business can do is build like a bigger company without actually hiring like one.

983
00:58:34.327 --> 00:58:38.011
But the problem we all face is that most companies don't have every skill in-house.

984
00:58:38.171 --> 00:58:44.477
So when I look at the businesses seeing real success today, the consistent pattern with all of them is how quickly they move.

985
00:58:44.577 --> 00:58:48.281
They bring in specialists with skills in emerging areas to keep themselves ahead.

986
00:58:48.461 --> 00:58:49.202
Even in our company,

987
00:58:49.362 --> 00:58:56.064
We spent the last year pulling in talent across areas like AI-native strategy, no-code builds, and product workflows.

988
00:58:56.204 --> 00:58:58.885
And we find this talent through our long-time partner, Fiverr Pro.

989
00:58:59.185 --> 00:59:12.710
Their premium service only shows you vetted talent, so you've always got the safeguard that anyone you pull in to help you with a complex project has the skills that you're after and will deliver to the same high standards as your internal team.

990
00:59:12.970 --> 00:59:14.891
And most importantly, they'll keep up with the pace.

991
00:59:15.011 --> 00:59:18.612
It's a simple strategy, but it lets us stay agile without compromising on quality.

992
00:59:18.832 --> 00:59:25.176
So if you need these kind of skills in your business, head to pro.fiverr.com to find pioneering talent to fill your business's gaps.

993
00:59:25.536 --> 00:59:27.217
That's pro.fiverr.com.

994
00:59:28.137 --> 00:59:30.779
You know the little traditional SIM card that goes inside of our phones?

995
00:59:31.119 --> 00:59:34.901
They haven't changed at all since they were invented in the 90s.

996
00:59:35.202 --> 00:59:44.707
You have this physical piece of plastic that means you're locked into one carrier, one network, and the second you cross a border, that carrier can start charging you whatever they want.

997
00:59:44.787 --> 00:59:46.068
But there are alternatives.

998
00:59:46.268 --> 00:59:48.190
And today's sponsor, Saylee, is one of them.

999
00:59:48.591 --> 00:59:55.018
It's an eSIM app that gives you a safe and secure data connection in over 200 destinations.

1000
00:59:55.218 --> 01:00:00.644
All of their eSIMs have built-in cybersecurity, which is great if you're traveling for work and looking at confidential material.

1001
01:00:01.025 --> 01:00:07.212
I've been using Saylee whenever I travel because the connection is always reliable and it saves me a ton of roaming fees.

1002
01:00:07.332 --> 01:00:12.176
It also means I don't have to deal with all of the faff that surrounds sorting out a sim everywhere I go.

1003
01:00:12.617 --> 01:00:17.501
If you want to give it a try, download the Saley app from the App Store now and scan the QR code on screen.

1004
01:00:17.862 --> 01:00:25.349
And if you want 15% off your first purchase, use my code DOAC when you get to checkout.

1005
01:00:25.609 --> 01:00:27.691
That's DOAC for 15% off.

1006
01:00:31.105 --> 01:00:33.928
My car that drives itself, that is AI technology.

1007
01:00:34.128 --> 01:00:34.328
Yes.

1008
01:00:34.949 --> 01:00:46.059
I sat here with Dara from Uber, and he was saying that, I think in a couple of years' time, we won't need drivers for Uber because the cars will drive themselves.

1009
01:00:46.119 --> 01:00:47.280
Like, they'll be fully autonomous.

1010
01:00:48.001 --> 01:00:49.662
And I think, if I'm not mistaken...

1011
01:00:50.812 --> 01:00:53.453
Driving is one of the biggest professions on planet Earth.

1012
01:00:54.193 --> 01:01:01.696
So when you hear people, when you hear the CEO saying that there will be job disruption, you say that they are not telling the truth.

1013
01:01:02.036 --> 01:01:02.236
Yes.

1014
01:01:02.676 --> 01:01:04.937
Or they're guessing in a way that's very good for them.

1015
01:01:05.117 --> 01:01:07.278
Think about it from the perspective of Microsoft.

1016
01:01:07.298 --> 01:01:07.938
Satya Nadella.

1017
01:01:08.618 --> 01:01:10.139
He's not going to be like, yeah, we're done after.

1018
01:01:10.159 --> 01:01:10.959
This is going to work, mate.

1019
01:01:11.279 --> 01:01:15.441
Of course he's going to talk his book and he's going to say, yeah, this is going to replace all workers.

1020
01:01:15.761 --> 01:01:16.682
It's going to be amazing.

1021
01:01:17.462 --> 01:01:18.522
It's going to be so powerful.

1022
01:01:18.582 --> 01:01:20.883
And then he'll change his tune and say, actually, it's not going to replace workers.

1023
01:01:20.923 --> 01:01:23.705
It'll make them more powerful because the things aren't catching up.

1024
01:01:23.945 --> 01:01:25.045
They're from Uber, for example.

1025
01:01:25.545 --> 01:01:30.888
Of course he's going to say, if this happens, then that would be good for Uber because Uber would just become...

1026
01:01:31.588 --> 01:01:32.990
An autonomous taxi service.

1027
01:01:33.391 --> 01:01:34.913
There's a reason that Waymo's taken...

1028
01:01:35.173 --> 01:01:36.215
I find Waymo fascinating.

1029
01:01:36.235 --> 01:01:37.196
I think that shit's really cool.

1030
01:01:37.236 --> 01:01:39.319
I think there are socioeconomic problems that will come from it.

1031
01:01:39.719 --> 01:01:42.183
I think there are actual real problems that will emerge.

1032
01:01:42.203 --> 01:01:43.404
And also... What kind of problems?

1033
01:01:43.445 --> 01:01:48.311
Well, I mean, socioeconomically, there are, like you said, one of the largest employment centers in the world.

1034
01:01:48.992 --> 01:01:50.813
I mean, just the economics of cabs would fall apart.

1035
01:01:50.833 --> 01:01:52.855
But again, we are nowhere, nowhere, nowhere near that.

1036
01:01:52.955 --> 01:01:53.875
We're not even close.

1037
01:01:54.195 --> 01:02:04.582
Waymo has had to do the smallest rollouts and the most controlled things because the problem with pretty much every AI system, but especially driving, is not the getting 95% of the way.

1038
01:02:04.882 --> 01:02:05.903
It's those edge cases.

1039
01:02:06.023 --> 01:02:08.565
It's raining, which is a big problem for them in San Francisco.

1040
01:02:08.985 --> 01:02:11.847
It's a kid runs across the road, but they're wearing a high-vis thing.

1041
01:02:11.867 --> 01:02:12.367
Yeah.

1042
01:02:12.487 --> 01:02:13.888
Does it even notice it's a child?

1043
01:02:14.008 --> 01:02:22.094
Again, this is a really interesting but very, very applicable example of the right comparison to be made shouldn't be autonomous vehicles versus perfection.

1044
01:02:22.555 --> 01:02:24.656
It should be autonomous vehicles versus human drivers.

1045
01:02:25.176 --> 01:02:29.400
I mean, I don't know if I agree because the human driver might make mistakes, sure, but...

1046
01:02:30.208 --> 01:02:32.050
Again, not an expert in autonomous cars.

1047
01:02:32.090 --> 01:02:32.911
Just want to be clear.

1048
01:02:33.451 --> 01:02:42.039
But if we're pushing autonomous cars out there willy-nilly, and we're not doing so in extremely controlled environments, those edge cases will multiply and be dangerous.

1049
01:02:42.319 --> 01:02:44.902
Yeah, they might be better at human drivers in some ways, but they might also...

1050
01:02:45.182 --> 01:02:50.567
I was in Vegas the other day, and I was in a hotel, and I watched a bunch of Zoox cars just get fucking stuck.

1051
01:02:50.847 --> 01:02:51.908
The autumnalous cars.

1052
01:02:51.928 --> 01:02:52.448
Yeah, yeah, yeah.

1053
01:02:52.508 --> 01:02:53.469
These weird boxy things.

1054
01:02:53.489 --> 01:02:55.150
And they just blocked the exit.

1055
01:02:55.550 --> 01:02:57.812
They just all kind of lined up and just fell asleep.

1056
01:02:58.372 --> 01:03:01.954
I saw the same thing actually happen outside of a hotel when it got out of a Waymo in San Francisco.

1057
01:03:01.974 --> 01:03:05.637
It just stopped at the... And then a bunch of cars and another Waymo got stuck behind it.

1058
01:03:05.937 --> 01:03:07.138
And these are kind of...

1059
01:03:07.218 --> 01:03:09.139
I've seen some human bad drivers as well.

1060
01:03:09.279 --> 01:03:09.759
I agree.

1061
01:03:09.859 --> 01:03:13.362
But it's just we have control over deploying these...

1062
01:03:13.942 --> 01:03:14.903
Bad or good drivers.

1063
01:03:15.163 --> 01:03:19.466
We have an ability to roll them out slowly, which is exactly what we should do.

1064
01:03:19.806 --> 01:03:21.308
I'm not saying autonomous cars are bad.

1065
01:03:21.628 --> 01:03:29.194
I'm saying we need to be so, so, so careful and treat them as guilty until proven innocent because we can prove...

1066
01:03:29.814 --> 01:03:31.596
And also, they have people overlooking them.

1067
01:03:31.616 --> 01:03:33.138
They actually have people monitoring the routes.

1068
01:03:33.498 --> 01:03:34.859
It is something they cannot rush out.

1069
01:03:35.059 --> 01:03:36.941
And it doesn't seem like they're rushing it, which is good.

1070
01:03:37.182 --> 01:03:38.483
And they're not promising the world.

1071
01:03:38.523 --> 01:03:38.863
I do agree.

1072
01:03:38.923 --> 01:03:43.708
Listen, I'm a big fan of taxi drivers generally, in part because I spend a lot of time in taxis.

1073
01:03:43.748 --> 01:03:48.353
And I think I'm not just getting in there because I want to get from A to B. I'm getting in there for lots of other reasons.

1074
01:03:49.014 --> 01:03:50.515
However, when I look at the stats...

1075
01:03:51.462 --> 01:03:56.766
around what is more dangerous, driving myself or having an autonomous vehicle drive me.

1076
01:03:57.266 --> 01:04:02.870
There's a 68% lower overall crash involvement rate when you're in an autonomous vehicle.

1077
01:04:03.511 --> 01:04:12.437
Autonomous vehicles experience roughly 2.1 police-reported crashes per million miles compared to humans that are at roughly 4.68 per million miles.

1078
01:04:12.537 --> 01:04:14.659
So a 55% reduction when you get in an autonomous vehicle.

1079
01:04:14.679 --> 01:04:14.939
Right.

1080
01:04:15.119 --> 01:04:21.841
An autonomous vehicle show an 80% to 81% reduction in crashes resulting in injuries versus human drivers.

1081
01:04:22.741 --> 01:04:32.564
So you're 85% less likely to be involved in a single vehicle crash, like hitting a wall or a tree, if you're an autonomous vehicle versus being driven by humans.

1082
01:04:32.744 --> 01:04:33.865
I agree, but...

1083
01:04:33.885 --> 01:04:34.385
So it's safer.

1084
01:04:35.981 --> 01:04:39.465
Also, that data is, what's the sample size of human drivers?

1085
01:04:39.525 --> 01:04:44.811
I mean, we've got many, many, many, many, many, many more years of drivers and many, many, many more years of accidents.

1086
01:04:44.971 --> 01:04:48.355
And also, man, does that not have anything to do with generative AI.

1087
01:04:48.956 --> 01:04:50.518
If we were just talking about that...

1088
01:04:50.878 --> 01:04:52.139
be having a different conversation.

1089
01:04:52.159 --> 01:04:54.240
I guess the question here was really around job disruption.

1090
01:04:54.280 --> 01:04:57.822
Like, you know, we look across industries and we go, driving's a massive profession.

1091
01:04:57.942 --> 01:05:00.783
Is there going to be job disruption because cars can now drive themselves?

1092
01:05:01.103 --> 01:05:11.769
If we think about white-collar, you know, jobs, you know, lawyers and accountants, people sit here and they tell me that lawyers and accountants, the profession, right, I should say some of the skills within the

1093
01:05:14.550 --> 01:05:42.192
ai's to do here's the thing lawyers for example great example always hearing fucking legal partners talking about ai never the associates the associates are the ones that go out and find the precedent they're the ones that go and do the grunt work they're the ones who pull emotions half the time the partner is the one that might be the litigant it may be the client facing but the ones that are actually doing the day-to-day work i'm not hearing from them i'm not hearing associates being like this is fucking awesome i'm hearing a bunch of well-paid people that

1094
01:05:44.234 --> 01:05:46.315
Yeah, I'm the greatest lawyer ever.

1095
01:05:46.675 --> 01:05:49.036
They're not the ones that I want to hear from, the actual workers.

1096
01:05:49.296 --> 01:05:51.297
White-collar labor disruption is not happening.

1097
01:05:51.457 --> 01:05:59.740
OpenAI had a study that came out, I think, like a week ago, that said there was no connection between spending on AI tokens and revenue per employee.

1098
01:06:00.320 --> 01:06:01.301
Like, this is open.

1099
01:06:01.321 --> 01:06:02.161
What does that mean?

1100
01:06:02.181 --> 01:06:02.821
Could you explain that to me?

1101
01:06:02.841 --> 01:06:07.363
As in, the more tokens you spend has no correlation at all.

1102
01:06:07.943 --> 01:06:09.204
with the amount of money you make.

1103
01:06:09.544 --> 01:06:10.824
It's the second report they've put out.

1104
01:06:11.025 --> 01:06:13.386
The other one was like, hallucinations are mathematically guaranteed.

1105
01:06:13.706 --> 01:06:16.447
Kind of almost, it's the one thing I respect about that company.

1106
01:06:16.527 --> 01:06:18.028
Occasionally, they just put out a study.

1107
01:06:18.048 --> 01:06:19.789
It's like, yeah, shit, it kind of sucks.

1108
01:06:20.810 --> 01:06:23.091
But the people that are having their lives disrupted,

1109
01:06:23.711 --> 01:06:25.792
work-wise, are art directors.

1110
01:06:25.852 --> 01:06:31.833
It's people, art directors, transcribers, translators, who have bosses that don't care about the output.

1111
01:06:32.173 --> 01:06:33.954
It's what they consider cheap work.

1112
01:06:34.254 --> 01:06:37.714
And the problem is, is those people would have automated your work away anyway.

1113
01:06:37.734 --> 01:06:38.475
They would have sold it.

1114
01:06:38.655 --> 01:06:40.255
They would have taken the cheapest offer.

1115
01:06:40.275 --> 01:06:41.655
They would have sold it to the global self.

1116
01:06:41.675 --> 01:06:43.376
They would have taken the shittiest option they could.

1117
01:06:43.756 --> 01:06:44.956
That is something that AI is doing.

1118
01:06:44.976 --> 01:06:47.857
And again, those people are not paying the actual cost of AI.

1119
01:06:47.897 --> 01:06:48.957
They're using a subscription.

1120
01:06:49.437 --> 01:06:51.658
The actual white-collar labor force

1121
01:06:52.685 --> 01:06:59.388
might have some things that are slightly changing, but there is no evidence of productivity gains.

1122
01:06:59.588 --> 01:07:02.970
In fact, if there were, they would be screaming it from the rooftops.

1123
01:07:02.990 --> 01:07:08.533
There was an Oxford Economics study last year where it's like, oh, young people are finding less jobs because of AI.

1124
01:07:08.753 --> 01:07:11.514
We actually read the study, which multiple journalists did not.

1125
01:07:11.974 --> 01:07:14.636
It was a single line that said, yeah, we saw some correlation.

1126
01:07:15.416 --> 01:07:16.317
Didn't give a number.

1127
01:07:16.877 --> 01:07:18.659
Didn't actually say what the correlation was.

1128
01:07:19.079 --> 01:07:29.088
We are so conditioned to believe that the rich and powerful know what they're doing that we internalize these narratives about like, well, previous booms lost a lot of money.

1129
01:07:29.188 --> 01:07:30.949
Well, technology takes time to do stuff.

1130
01:07:31.350 --> 01:07:34.993
And they are intentionally playing on those mythologies.

1131
01:07:35.013 --> 01:07:39.357
They are playing on these knowing that journalists, analysts, investors...

1132
01:07:40.057 --> 01:07:40.838
We'll believe them.

1133
01:07:41.239 --> 01:07:44.883
And this is partly because our realities are defined by stock prices.

1134
01:07:44.984 --> 01:07:49.550
Because the stock prices of these companies went up, we're like, oh, look, it must be working, right?

1135
01:07:49.990 --> 01:07:51.712
Both of those things you said were true, though, right?

1136
01:07:51.752 --> 01:07:54.376
Like that previous technologies didn't make money at the start.

1137
01:07:54.496 --> 01:07:57.260
And the other one you said was, they'll get better.

1138
01:07:57.660 --> 01:07:58.220
But that's the thing.

1139
01:07:58.660 --> 01:08:01.441
Okay, because another thing got better, this will get better?

1140
01:08:01.581 --> 01:08:05.122
No, but there's got to be something that they're saying that is fundamentally not true.

1141
01:08:05.142 --> 01:08:08.063
Because those are two true statements that, okay, technology often starts...

1142
01:08:08.083 --> 01:08:08.523
Okay, I know.

1143
01:08:08.563 --> 01:08:09.204
I get what you mean.

1144
01:08:09.284 --> 01:08:09.504
Yeah.

1145
01:08:09.604 --> 01:08:13.425
What they are fundamentally misleading people about is how possible it is.

1146
01:08:13.765 --> 01:08:16.286
How many actual signs they have, because they don't have the signs.

1147
01:08:16.346 --> 01:08:32.512
If they had the signs, as in the signs of this getting cheaper, as in the signs of this being able to autonomously do work without the Rube Goldberg machine, and even then, in a reliable way that was making the customer more money, being productive in a way you can say with your whole chest without a series of asterisks.

1148
01:08:33.172 --> 01:08:35.033
And that's how it is across the board.

1149
01:08:35.073 --> 01:08:35.633
The people...

1150
01:08:36.493 --> 01:08:42.680
that are most excited about this, psychopaths on Twitter in many cases, are people that I believe... Psychopaths on Twitter.

1151
01:08:42.900 --> 01:08:43.821
They really are some...

1152
01:08:44.101 --> 01:08:49.787
I'm sorry, there are some people on Twitter... Because the other thing about this is, this is really unique to the AI industry.

1153
01:08:49.807 --> 01:08:53.150
I've never seen it any other industry outside of maybe, like, sports teams.

1154
01:08:53.771 --> 01:08:58.076
The attachment that some people online have to these companies, if you dare...

1155
01:08:58.536 --> 01:09:27.219
dare to criticize anthropic it's almost this religious attachment good example was this week bloomberg reported that open ai was on track to hit 40 billion dollars in annualized revenue month times 12 four weeks times 13 we don't know they don't define it i saw multiple people and i going actually it's 60 billion it's actually 60 billion i heard from someone it is like a cult and it's a cult of software driven around growth and this idea that by backing the right horse you

1156
01:09:27.679 --> 01:09:29.841
you will have some grand thing.

1157
01:09:30.001 --> 01:09:35.726
And OpenAI in particular, in particular Mr. Altman, they have been fomenting this.

1158
01:09:35.766 --> 01:09:40.210
They're Tibo as well, the TIBO, one of the guys at OpenAI.

1159
01:09:40.510 --> 01:09:41.551
They foment this thing online.

1160
01:09:41.571 --> 01:09:47.537
They build this kind of parasocial relationship with both the large language model themselves and the companies.

1161
01:09:48.157 --> 01:09:51.200
And one's allegiance to the companies is so important.

1162
01:09:51.580 --> 01:09:52.622
It's truly vile.

1163
01:09:53.062 --> 01:10:02.277
If only these people gave a fuck about, I don't know, Medicare for All or poverty or things like actual problems in the world versus are we buying enough GPUs?

1164
01:10:02.297 --> 01:10:05.342
Do you know what's interesting is some of what your narrative...

1165
01:10:06.730 --> 01:10:08.693
one would argue, actually helps them.

1166
01:10:09.114 --> 01:10:09.314
How?

1167
01:10:09.895 --> 01:10:21.712
Because, you know, the AI doomers that have come here and told, you know, some of the original founding fathers of AI, like Geoffrey Hinton, have told me that what they're building is highly, highly dangerous and that it will be fundamentally disruptive to society.

1168
01:10:22.433 --> 01:10:22.633
And...

1169
01:10:23.514 --> 01:10:29.724
It's interesting because some of the CEOs who you've mentioned, their historical narrative was also, by the way, this is really fucking dangerous.

1170
01:10:29.764 --> 01:10:29.984
Yeah.

1171
01:10:30.024 --> 01:10:32.828
And there is a significant chance we could fuck up the planet.

1172
01:10:32.969 --> 01:10:36.294
And what we've seen is this slow pivot away from it.

1173
01:10:36.314 --> 01:10:36.794
Which is so strange.

1174
01:10:36.815 --> 01:10:38.778
Because now they're getting booed and they're being attacked.

1175
01:10:38.798 --> 01:10:38.978
Yeah.

1176
01:10:39.118 --> 01:10:40.798
They've been this slow pivot away from it.

1177
01:10:41.259 --> 01:10:45.279
And the pivot almost sounds a little bit like your narrative.

1178
01:10:45.780 --> 01:10:48.260
It now sounds like, actually, no, it's not going to change anything.

1179
01:10:48.520 --> 01:10:49.240
And you're all going to be fine.

1180
01:10:49.260 --> 01:10:51.661
And it's now, it's just not, it's not dangerous at all.

1181
01:10:51.761 --> 01:10:52.701
But that's the funny thing.

1182
01:10:53.221 --> 01:10:59.643
And that's why I'm saying like, I actually think there might be a couple of PR people at these big AI companies thinking, thank God for Ed.

1183
01:11:00.143 --> 01:11:01.884
Oh, I don't know about that, mate.

1184
01:11:02.145 --> 01:11:02.465
Some of it.

1185
01:11:02.725 --> 01:11:04.987
Because you're saying, actually, don't worry.

1186
01:11:05.027 --> 01:11:06.027
Everything's going to be fine.

1187
01:11:06.087 --> 01:11:07.108
It's not going to take your job.

1188
01:11:07.148 --> 01:11:08.129
It's not going to disrupt the economy.

1189
01:11:08.149 --> 01:11:08.729
It's just a fad.

1190
01:11:08.749 --> 01:11:09.250
There's no technology.

1191
01:11:09.690 --> 01:11:11.131
And I think they don't think that.

1192
01:11:11.431 --> 01:11:11.892
Here's the thing.

1193
01:11:12.212 --> 01:11:15.955
I think Altman and Amadeus are some of the most deeply corrupt and cynical people in the world.

1194
01:11:16.555 --> 01:11:22.560
Of course they were going to say, it was early 2023, Altman said, we're a little bit scared about what we're creating.

1195
01:11:22.580 --> 01:11:22.880
Right.

1196
01:11:23.320 --> 01:11:42.063
oh shut up i'm just i hear that and i feel so frustrated because i've met so many of these rich fucking liars these people and you know why he wants to say that so you'll invest in his company and buy the software so you'll be scared that if you don't use ai today you'll be left behind in the future which is their continual narrative that if you don't get on the train today

1197
01:11:43.637 --> 01:11:44.397
you'll be left behind.

1198
01:11:44.477 --> 01:11:48.259
By the way, every single scam and con starts with rushing you.

1199
01:11:48.719 --> 01:11:52.480
Every single trick in history begins with saying, you must do this now.

1200
01:11:52.820 --> 01:12:01.703
And best piece of advice I ever got was, if anyone tries to rush you, and it's not literally a mortal thing like you are bleeding or on fire or the house is on fire, slow down.

1201
01:12:02.083 --> 01:12:04.724
And yet all of these companies say, so scary.

1202
01:12:04.824 --> 01:12:06.545
And now they're talking about slowdowns, but you're...

1203
01:12:06.585 --> 01:12:10.228
ever noticed that Amaday and Altman, they say, oh, maybe we should slow down progress.

1204
01:12:10.428 --> 01:12:11.108
And then they don't.

1205
01:12:11.669 --> 01:12:14.691
Right now, Altman's saying, oh, we've slowed down progress because we're so delayed.

1206
01:12:14.811 --> 01:12:15.572
Now they're out of compute.

1207
01:12:16.032 --> 01:12:16.772
Now they're doing it.

1208
01:12:16.893 --> 01:12:20.175
I can guarantee you, by the way, their PR people do not like me.

1209
01:12:20.295 --> 01:12:23.117
I don't think OpenAI's PR people are super fond of me.

1210
01:12:23.137 --> 01:12:26.319
But I bet there's elements of what you're saying, because you're calming people.

1211
01:12:26.660 --> 01:12:28.781
You are theoretically calming down the general public.

1212
01:12:28.821 --> 01:12:29.482
And you know what?

1213
01:12:29.702 --> 01:12:30.683
I hope I am, because...

1214
01:12:31.363 --> 01:12:33.145
The fear-based tactics is horrible.

1215
01:12:33.285 --> 01:12:34.326
These companies don't want that.

1216
01:12:34.366 --> 01:12:35.647
These companies want people scared.

1217
01:12:35.807 --> 01:12:38.330
I'm 100% sure.

1218
01:12:38.410 --> 01:12:39.551
I just fundamentally disagree.

1219
01:12:39.591 --> 01:12:40.371
I think it...

1220
01:12:40.612 --> 01:12:45.076
So the timelines, and I sit here, and what I do is I log their quotes over time.

1221
01:12:45.276 --> 01:12:45.416
Oh.

1222
01:12:45.516 --> 01:12:46.918
And I read them out from 2015 to 2026.

1223
01:12:46.958 --> 01:12:48.399
And the change you see...

1224
01:12:51.822 --> 01:12:54.383
is them going from, there could be extinction.

1225
01:12:54.623 --> 01:12:55.883
That's the narrative, the early narrative.

1226
01:12:55.983 --> 01:12:56.723
Elon said it himself.

1227
01:12:56.743 --> 01:12:59.424
He says it's the single most dangerous thing in the- Elon says a lot of things.

1228
01:12:59.444 --> 01:13:03.685
And then you track it over time and it evolves to this age of abundance.

1229
01:13:03.705 --> 01:13:05.326
We're all going to have unlimited stuff.

1230
01:13:05.786 --> 01:13:09.987
And then the new slogan at Track TVT is intelligence for everyone.

1231
01:13:10.608 --> 01:13:15.469
It's suddenly, and all the, and whenever Dario comes out and says, by the way, it's really rockin' dangerous.

1232
01:13:16.957 --> 01:13:17.778
Yeah, they hate him.

1233
01:13:18.619 --> 01:13:21.662
They're like, Dario, shut the fuck up.

1234
01:13:21.702 --> 01:13:24.686
Honestly, I've been saying Dario, shut the fuck up for years.

1235
01:13:24.846 --> 01:13:33.556
But the thing is, I get your point where it's like, I don't think they've changed to calm the public down so much as they're desperate to not get regulated, which is laughable.

1236
01:13:33.736 --> 01:13:34.877
We don't regulate tech.

1237
01:13:35.378 --> 01:13:36.539
We don't regulate shit.

1238
01:13:36.739 --> 01:13:38.981
America doesn't regulate shit.

1239
01:13:39.521 --> 01:13:45.025
We are still trapped in the hands of Milton Friedman, Margaret Thatcher, and fucking Ronald Reagan.

1240
01:13:45.245 --> 01:13:51.469
We're still stuck in the neoliberalistic hellscape, which is growth at all costs, free market capitalism.

1241
01:13:51.709 --> 01:13:53.391
So no, no one's regulating.

1242
01:13:53.451 --> 01:13:57.213
The regulation of these companies should have been, I don't know, breaking up.

1243
01:13:57.614 --> 01:13:58.594
Put these bastards aside.

1244
01:13:58.934 --> 01:14:00.355
Break up these fuckers, for sure.

1245
01:14:00.416 --> 01:14:01.616
We shouldn't have companies this big.

1246
01:14:01.636 --> 01:14:02.337
It makes things worse.

1247
01:14:02.357 --> 01:14:03.738
But these technologies are dangerous.

1248
01:14:04.763 --> 01:14:07.047
I mean, they're dangerous, but not in the ways they've been warning about.

1249
01:14:07.888 --> 01:14:09.190
If we think about cyber hacking.

1250
01:14:09.831 --> 01:14:10.071
Right.

1251
01:14:10.131 --> 01:14:13.837
And just be clear, those cyber hacking things that happened were not a result of...

1252
01:14:13.857 --> 01:14:15.960
They were like, break out of the sandbox.

1253
01:14:15.980 --> 01:14:17.422
And then they set the sandbox up.

1254
01:14:18.328 --> 01:14:18.548
wrong.

1255
01:14:18.668 --> 01:14:20.189
They set up the server they were on wrong.

1256
01:14:20.429 --> 01:14:41.495
But I mean, you know, advanced AI models could very easily, because they can go out onto the open internet as agents, they could very easily go and look at code bases of different websites, find vulnerabilities and exploit those vulnerabilities at scale and arguably at a higher intelligence and faster and wider than a human hacker could, theoretically.

1257
01:14:41.515 --> 01:14:42.275
So that's dangerous.

1258
01:14:42.516 --> 01:14:43.516
Well, here's the funny thing.

1259
01:14:44.383 --> 01:14:49.047
We don't know how much compute was spent to do the hugging face attack, the open AI one.

1260
01:14:49.167 --> 01:14:51.929
We also do know that they improperly set up the server to keep it in.

1261
01:14:52.109 --> 01:14:54.371
They thought they'd turn the internet off, and they didn't.

1262
01:14:54.911 --> 01:15:00.616
That's human error, and that's human error in a sense that, yeah, they threw about an indeterminately large amount of compute.

1263
01:15:01.076 --> 01:15:06.561
This is dangerous, but people keep saying, we can't let the Chinese get a hold of these models.

1264
01:15:06.601 --> 01:15:09.743
We couldn't possibly, because what if these models fall into the wrong hands?

1265
01:15:10.004 --> 01:15:11.485
They're already in the wrong hands.

1266
01:15:11.865 --> 01:15:14.646
Mark Zuckerberg, Sam Altman, Dario Amadei.

1267
01:15:14.866 --> 01:15:18.227
The wrong hands are the hands of those who are running these companies.

1268
01:15:18.547 --> 01:15:21.868
We should not be training these models to do these things.

1269
01:15:22.028 --> 01:15:23.308
I don't know why the fuck we're doing it.

1270
01:15:23.749 --> 01:15:26.489
Other than they've run out of other things they can train on.

1271
01:15:26.529 --> 01:15:27.049
There's a ton.

1272
01:15:27.410 --> 01:15:29.470
And the fact that they can do it, it's kind of interesting.

1273
01:15:29.630 --> 01:15:31.651
But would you agree that it's...

1274
01:15:32.451 --> 01:15:45.197
An intelligence, and I'll call it that, you might disagree with that terminology, but an intelligence that can go out onto the internet and click around and take actions is inherently, there's risks associated with that.

1275
01:15:45.457 --> 01:15:47.778
Well, the second part I agree with, the risks.

1276
01:15:48.318 --> 01:15:51.420
We've had people running automated scripts and hacking scripts for a while.

1277
01:15:51.440 --> 01:15:53.401
We've had hackers doing that for years and years and years.

1278
01:15:53.881 --> 01:15:56.062
This is brute forcing it with a bunch of compute and yet...

1279
01:15:56.782 --> 01:15:57.403
It is dangerous.

1280
01:15:57.663 --> 01:15:59.584
These companies are doing something dangerous.

1281
01:16:00.105 --> 01:16:02.707
That is not what Jeffrey Hinton et al.

1282
01:16:02.987 --> 01:16:03.688
have been warning about.

1283
01:16:03.708 --> 01:16:06.170
They've been saying, oh, these things could destroy society.

1284
01:16:06.190 --> 01:16:07.411
They could manipulate people.

1285
01:16:07.651 --> 01:16:09.412
When you actually look at the underlying things, not so much.

1286
01:16:09.913 --> 01:16:12.695
Jeffrey Hinton as well, talking his book, still got his Google stock, I think.

1287
01:16:13.035 --> 01:16:20.219
Weirdly enough, he left Google because he was worried about the AI there, but then immediately made a comment being like, yeah, actually, though, Google's very responsible.

1288
01:16:20.339 --> 01:16:21.039
Strange thing, that.

1289
01:16:21.600 --> 01:16:23.461
But let's get back to the cybersecurity side.

1290
01:16:23.681 --> 01:16:24.061
I agree.

1291
01:16:24.181 --> 01:16:24.922
This is dangerous.

1292
01:16:25.222 --> 01:16:27.263
These people should not have access to so much compute.

1293
01:16:27.283 --> 01:16:28.604
They clearly don't know what to do with it.

1294
01:16:29.184 --> 01:16:31.045
There's a really easy way of dealing with this.

1295
01:16:31.285 --> 01:16:32.726
It's not letting them use so much compute.

1296
01:16:33.006 --> 01:16:35.147
It's regulating that part out of existence.

1297
01:16:35.307 --> 01:16:36.348
What if the Chinese do it?

1298
01:16:37.208 --> 01:16:38.729
The Chinese are able to distill the models.

1299
01:16:39.630 --> 01:16:39.990
And also,

1300
01:16:41.495 --> 01:16:43.856
I don't know, regulate it and stop.

1301
01:16:44.276 --> 01:16:52.140
I feel like with this particular thing as well, we got to this point and let the genie out of the bottle, to use an annoying Sam Altman term.

1302
01:16:53.020 --> 01:16:57.382
We let this happen because we let these companies be unregulated and use as much computers as we want.

1303
01:16:57.402 --> 01:17:00.964
We had these fucking enablers allow them to burn as much computers as they want.

1304
01:17:01.324 --> 01:17:07.867
And also, for all of these dire warnings about AI dangers, no one seems to have fucking done anything.

1305
01:17:08.227 --> 01:17:09.308
Okay, we're going to play a game, Ed.

1306
01:17:09.408 --> 01:17:09.869
Let's play it.

1307
01:17:09.949 --> 01:17:16.054
On these cards here, I have the things that you consider to be myths about the AI industry.

1308
01:17:16.894 --> 01:17:22.119
The challenge is I want you to give me one sentence on each myth.

1309
01:17:22.639 --> 01:17:23.040
Oh, Christ.

1310
01:17:23.060 --> 01:17:24.121
So just your first reaction.

1311
01:17:24.141 --> 01:17:29.805
You're going to pick it up, you're going to read it, and then you're going to give me one sentence on your opinion of that belief.

1312
01:17:30.106 --> 01:17:30.646
Okay, let's go.

1313
01:17:31.086 --> 01:17:31.647
Let's do this.

1314
01:17:36.592 --> 01:17:37.112
What does it say?

1315
01:17:37.152 --> 01:17:38.272
And what's your one sentence?

1316
01:17:38.292 --> 01:17:41.633
It says the AI industry is creating enormous economic growth.

1317
01:17:42.354 --> 01:17:42.934
No, it's not.

1318
01:17:42.994 --> 01:17:44.014
It's nowhere in the data.

1319
01:17:45.294 --> 01:17:46.215
Okay.

1320
01:17:46.255 --> 01:17:48.055
It's just, may I do a second sentence?

1321
01:17:48.075 --> 01:17:48.415
Go ahead.

1322
01:17:49.276 --> 01:17:56.958
Pretty much all of the economics is either NVIDIA feeding money to companies like CoreWeave or these three companies feeding money to these ones to spend it with them.

1323
01:17:58.361 --> 01:17:58.561
Okay.

1324
01:17:59.161 --> 01:18:02.842
And what evidence do you have that it's not causing economic growth?

1325
01:18:03.202 --> 01:18:10.424
Just to be clear, other than the spend on semiconductors, so the speculative investment in GPUs and data center infrastructure, that's happening.

1326
01:18:10.884 --> 01:18:21.007
But as far as like spend on AI goes, barely cracking $100 billion, and most of that is just these two running their services and paying these three companies, Oracle, Core, even others.

1327
01:18:21.207 --> 01:18:25.268
But $100 billion is a lot of money for a relatively new technology.

1328
01:18:25.308 --> 01:18:33.469
Not when you've spent $300 billion in equity funding and if we're going with just these three, I think $600 billion in capital expenditures.

1329
01:18:33.669 --> 01:18:34.270
Yeah, I get that.

1330
01:18:34.290 --> 01:18:35.230
That means it's not profitable.

1331
01:18:35.350 --> 01:18:38.510
But the $100 billion is an expression of consumer demand.

1332
01:18:38.791 --> 01:18:42.551
When the compute is mostly driven by subscriptions that are subsidized, no, it's not.

1333
01:18:42.591 --> 01:18:46.572
When you're giving someone $20 or $40 for a dollar, they are going to use it more.

1334
01:18:46.832 --> 01:18:50.213
If this was all on a per million token basis, we'd be having a different conversation.

1335
01:18:50.233 --> 01:18:50.453
Okay, fair.

1336
01:18:50.873 --> 01:18:51.033
Fine.

1337
01:18:51.373 --> 01:18:51.533
Cool.

1338
01:18:51.753 --> 01:18:52.034
Next one.

1339
01:18:53.875 --> 01:18:57.237
The United States need to spend trillions to beat China in the AI race.

1340
01:18:59.179 --> 01:18:59.659
Let's see.

1341
01:19:01.000 --> 01:19:01.800
What AI race?

1342
01:19:03.633 --> 01:19:05.194
that's actually my point.

1343
01:19:05.514 --> 01:19:06.974
It's what AI race is there?

1344
01:19:07.254 --> 01:19:09.335
Is it to make big, scary LLMs?

1345
01:19:09.475 --> 01:19:10.495
They did that already.

1346
01:19:10.815 --> 01:19:13.856
Without the NVIDIA GPUs, by the way, they've got Blackwell GPUs.

1347
01:19:13.896 --> 01:19:16.377
Kakashi and Justario, two amazing analysts I love.

1348
01:19:16.797 --> 01:19:17.778
They've been on this for years.

1349
01:19:17.818 --> 01:19:21.899
It's like, China's already had NVIDIA GPUs that they're not meant to have for years.

1350
01:19:22.059 --> 01:19:23.479
But also, to do what?

1351
01:19:23.499 --> 01:19:25.020
They already got the LLMs.

1352
01:19:25.360 --> 01:19:26.180
What's the race to do?

1353
01:19:26.240 --> 01:19:27.481
To make us spend more money than them?

1354
01:19:27.721 --> 01:19:30.141
For us to constantly piss our pants worrying about China?

1355
01:19:30.441 --> 01:19:32.142
Because they won, if that's the case.

1356
01:19:33.617 --> 01:19:34.297
Myth number three.

1357
01:19:34.877 --> 01:19:36.898
AI will replace all human jobs.

1358
01:19:38.038 --> 01:19:39.318
That just isn't happening.

1359
01:19:39.358 --> 01:19:41.179
There's no economic data to support it.

1360
01:19:42.279 --> 01:19:43.759
Will it replace some jobs?

1361
01:19:44.479 --> 01:19:49.320
I mean, it's replaced some contract labor that would otherwise be replaced with cheap labor out in the global south.

1362
01:19:49.940 --> 01:19:51.881
It's a digital globalization in that sense.

1363
01:19:51.941 --> 01:19:56.122
But all jobs, most jobs, a lot of jobs, no.

1364
01:19:56.282 --> 01:19:57.222
What about robotics?

1365
01:19:57.642 --> 01:19:59.402
Robotics is not what we're talking about.

1366
01:19:59.542 --> 01:20:00.663
Robotics is a very different thing.

1367
01:20:00.683 --> 01:20:02.363
And even then... Robotics will be powered by AI.

1368
01:20:03.083 --> 01:20:05.285
I mean, yes, but there are tons of different kinds of AI.

1369
01:20:05.325 --> 01:20:07.186
We're talking explicitly about generative AI.

1370
01:20:07.366 --> 01:20:07.887
And that's what I...

1371
01:20:08.107 --> 01:20:09.508
This is from my Mythbusters piece.

1372
01:20:09.568 --> 01:20:11.109
That was definitely about generative AI.

1373
01:20:11.169 --> 01:20:12.710
Okay, but what about robotics?

1374
01:20:12.730 --> 01:20:13.231
The thing is...

1375
01:20:13.251 --> 01:20:16.633
The Optimus robot that... Elon's working on it, Tesla.

1376
01:20:16.653 --> 01:20:21.597
The one where even in the demo of The Hand, he did like they had to have a guy controlling it.

1377
01:20:21.617 --> 01:20:22.678
It wasn't doing an autonomous thing.

1378
01:20:22.758 --> 01:20:23.238
Here's the thing.

1379
01:20:23.719 --> 01:20:24.599
If they can...

1380
01:20:25.612 --> 01:20:27.574
beat all these challenges, yeah, robotics would be really cool.

1381
01:20:27.934 --> 01:20:33.178
I don't know how long that's... That's one I'd actually be willing to believe in a couple decades.

1382
01:20:33.539 --> 01:20:35.160
Have you seen them Chinese robots?

1383
01:20:35.220 --> 01:20:35.981
I know you've seen them Chinese.

1384
01:20:36.001 --> 01:20:37.062
Well, the Unie, what's it called?

1385
01:20:37.602 --> 01:20:40.464
The one that can dance and that, but they can't really do human things?

1386
01:20:41.285 --> 01:20:43.167
Well, it's just, it is pretty mind-blowing.

1387
01:20:44.047 --> 01:20:45.228
Robotics are fucking cool.

1388
01:20:45.669 --> 01:20:47.950
Like, I'm not going to pretend I don't think robots are cool.

1389
01:20:48.111 --> 01:20:51.193
I wish they were building robots and actually doing cool shit.

1390
01:20:51.253 --> 01:20:54.396
I wish the tech industry still made fun stuff and interesting stuff.

1391
01:20:54.496 --> 01:20:56.477
Instead, we get these fucking large language models.

1392
01:20:56.818 --> 01:21:02.923
But AI plus robotics is, you know, I was in San Francisco and I went to this massive incubator there.

1393
01:21:03.503 --> 01:21:06.204
And when I'd gone there three years earlier, it was all software startups.

1394
01:21:06.284 --> 01:21:06.484
Right.

1395
01:21:06.704 --> 01:21:10.085
And when I went back three years later, it was all these robot startups.

1396
01:21:10.105 --> 01:21:13.606
And I remember saying to the founder of the incubator, I was like, why is everything robots now?

1397
01:21:13.846 --> 01:21:17.507
There was this one robot where it was just the arm and it had a frying pan on it.

1398
01:21:17.667 --> 01:21:17.888
Yeah.

1399
01:21:17.988 --> 01:21:19.328
And its whole thing is it cooks for you.

1400
01:21:19.548 --> 01:21:19.808
Yeah.

1401
01:21:19.928 --> 01:21:21.969
So it was showing me it cooking, whatever.

1402
01:21:22.270 --> 01:21:23.470
And he goes, well, you know the arm?

1403
01:21:23.590 --> 01:21:28.133
He goes, the hardware part, the physical parts, that's always been fairly cheap.

1404
01:21:28.213 --> 01:21:28.473
Yeah.

1405
01:21:28.693 --> 01:21:30.894
He goes, the expensive part was the intelligence.

1406
01:21:30.914 --> 01:21:31.175
Yeah.

1407
01:21:31.235 --> 01:21:32.575
And now that's come down to pennies.

1408
01:21:32.695 --> 01:21:38.019
So what you're seeing is this explosion in the robotics industry, because robotics is a function of intelligence plus hardware.

1409
01:21:38.199 --> 01:21:38.959
We've always had the hardware.

1410
01:21:38.979 --> 01:21:40.280
And a ton of data, though, as well.

1411
01:21:40.320 --> 01:21:40.440
Yeah.

1412
01:21:40.460 --> 01:21:42.181
And the data is very expensive.

1413
01:21:42.561 --> 01:21:42.801
Yeah.

1414
01:21:43.042 --> 01:21:45.884
The thing is, cybercabs rolled out real slow.

1415
01:21:46.345 --> 01:21:47.406
It's going to take a long time.

1416
01:21:47.686 --> 01:21:48.527
It could be a threat.

1417
01:21:48.807 --> 01:21:51.970
If they do a robot that could replace a human job, sure it could.

1418
01:21:52.370 --> 01:21:54.412
But human jobs are multifaceted.

1419
01:21:54.792 --> 01:21:56.154
Human jobs change with environments.

1420
01:21:56.194 --> 01:22:01.278
And also, a lot of human jobs that you might think of like, I don't know, dishwashing robot, for example.

1421
01:22:01.439 --> 01:22:01.639
Yeah.

1422
01:22:02.481 --> 01:22:08.230
Some guy at a restaurant isn't paying 10, 20 grand for a robot to replace the job that they're already not paying enough for.

1423
01:22:08.891 --> 01:22:12.035
The point is, yeah, it could if you can replace the jobs.

1424
01:22:12.616 --> 01:22:14.179
That is not what we're talking about with this.

1425
01:22:14.510 --> 01:22:18.152
Yeah, I ask these questions not because I'm trying to be like devil.

1426
01:22:18.472 --> 01:22:20.874
Actually, I'm trying to form my own opinion on these things.

1427
01:22:21.134 --> 01:22:28.018
And I do think, you know, as it's written there, it says AI will replace all human jobs.

1428
01:22:28.218 --> 01:22:28.658
Obviously not.

1429
01:22:28.938 --> 01:22:29.639
Obviously, that's bullshit.

1430
01:22:30.139 --> 01:22:38.564
But I'm trying to figure out if the truth is somewhere in the middle, that there's a certain type of job, which actually humans probably shouldn't have ever been doing, really, that's...

1431
01:22:38.924 --> 01:23:03.975
if you think back through history there was someone's job just to sit in an elevator and press the buttons right that's an example of a job that humans probably shouldn't have been doing and as technology gets more advanced it takes on a lot of that sort of automated monotonous stuff right the thing is with this particular thing that i know that this is from it's a specific blog i wrote i was explicitly talking about generative ai though i was explicitly talking about people when they say this they are referring to that so you're not talking about

1432
01:23:04.035 --> 01:23:27.990
agentic ai which agentic ai is llms agentic ai is just a fancy way of saying an llm talking to another llm with a harness on top that is still a lamp's agentic ai is one of the the bigger lies they tell it's like when you hear agent you're meant to think autonomous ai can do what you want it's still llms it's still llms talking to other llms taking screenshots and putting them in llms oh god yeah okay but but you know i could i could make the case that

1433
01:23:29.766 --> 01:23:31.267
I'm just thinking about my personal usage.

1434
01:23:31.587 --> 01:23:35.648
I definitely use agents to do things that I would have previously asked people to do.

1435
01:23:35.668 --> 01:23:38.189
It's not to say that I still don't hire because we're hiring like crazy.

1436
01:23:38.529 --> 01:23:38.769
Yeah.

1437
01:23:38.990 --> 01:23:42.871
And I still, in that particular function, I'm thinking about like the chief of staff role.

1438
01:23:43.591 --> 01:23:48.733
So my chief of staff would have triaged all of my inboxes previously and put them somewhere and told me about them.

1439
01:23:48.793 --> 01:23:51.514
Or maybe once upon a time, show me a piece of paper back in the day, I guess.

1440
01:23:52.015 --> 01:23:54.055
Now my chief of staff is no longer doing that job.

1441
01:23:54.255 --> 01:23:55.696
You still have a chief of staff, though.

1442
01:23:56.156 --> 01:23:56.638
This is what I'm saying.

1443
01:23:56.658 --> 01:23:57.680
They're doing other things.

1444
01:23:58.121 --> 01:23:58.482
Right.

1445
01:23:58.602 --> 01:24:00.947
But the thing is, again, what you were describing is...

1446
01:24:02.487 --> 01:24:03.507
Fairly basic automation.

1447
01:24:03.527 --> 01:24:04.607
I don't know what the tasks are.

1448
01:24:04.627 --> 01:24:05.808
Yeah, it's fairly basic email.

1449
01:24:06.288 --> 01:24:08.308
Didn't spend a trillion dollars on triaging email.

1450
01:24:08.428 --> 01:24:09.689
Like, that's the promise.

1451
01:24:09.809 --> 01:24:14.409
If they'd spent $10 billion and this was much smaller and you're like, I go, cool, software, yay.

1452
01:24:14.429 --> 01:24:18.630
A lot of the things that people are impressed with, like script stuff as well, it's just LLM's doing Python.

1453
01:24:18.850 --> 01:24:20.311
You should be impressed by Python code.

1454
01:24:20.331 --> 01:24:21.251
Python's incredible.

1455
01:24:21.271 --> 01:24:22.411
You can scrape websites.

1456
01:24:22.431 --> 01:24:23.151
You can download shit.

1457
01:24:23.171 --> 01:24:23.711
It's awesome.

1458
01:24:24.312 --> 01:24:25.732
But the point I'm making is...

1459
01:24:26.592 --> 01:24:32.494
None of this would be anywhere near as much of a problem if they didn't ask for all of the attention, all of the money and promise the world.

1460
01:24:33.034 --> 01:24:34.455
It's their promises that are the problem.

1461
01:24:34.535 --> 01:24:42.898
And the journalists who went along with it and the analysts and the Twitter people who went along with this saying that this would change everything and replace everything and leaving the realm of reality.

1462
01:24:43.318 --> 01:24:48.820
Is there any technological innovation through history that was really, really game changing where that didn't happen?

1463
01:24:50.860 --> 01:24:53.361
I mean, the internet...

1464
01:24:53.882 --> 01:24:55.182
I mean, people overpromised that.

1465
01:24:55.382 --> 01:25:00.585
I mean, they overpromised on the businesses, but I've read through a great many pieces about the early internet.

1466
01:25:01.125 --> 01:25:04.587
A lot of people were excited but hesitant.

1467
01:25:04.807 --> 01:25:11.210
They were worried that there was not enough demand, but they were still like, oh yeah, this could have potential ramifications if it happened.

1468
01:25:11.630 --> 01:25:12.291
People were not...

1469
01:25:12.951 --> 01:25:14.973
super negative about the internet.

1470
01:25:15.214 --> 01:25:17.996
A lot of the skeptics were saying we're worried about an overload of bad information.

1471
01:25:18.537 --> 01:25:19.318
Look at where we are.

1472
01:25:19.718 --> 01:25:24.323
A lot of people were worried about the social consequences of everyone talking online, which they were correct about.

1473
01:25:24.783 --> 01:25:31.090
With the economic things, they were specifically talking about like the globe, which I think made hundreds of thousands of dollars and had like a...

1474
01:25:31.390 --> 01:25:33.411
I think a billion-dollar market capper.

1475
01:25:33.431 --> 01:25:35.973
Yeah, there was massive hype in the dot-com era.

1476
01:25:36.133 --> 01:25:37.354
I read a lot of those stories.

1477
01:25:37.374 --> 01:25:38.395
The hype was nowhere near.

1478
01:25:38.515 --> 01:25:43.058
You didn't have articles everywhere that were saying, if you don't get online, you'll be left behind.

1479
01:25:43.118 --> 01:25:45.920
You didn't have professional consequences.

1480
01:25:45.940 --> 01:25:47.541
Nick Suresh mentioned his blog earlier.

1481
01:25:47.961 --> 01:25:52.844
He described this thing, AI's of history and global decision-making, where he said that you have...

1482
01:25:53.725 --> 01:26:08.957
businesses you work at where if you don't say that you're more productive with AI whether or not it's true is irrelevant you have professional consequences you can get fired there are people having to AI wash their jobs by saying AI did it otherwise they're bosses who don't do shit

1483
01:26:09.858 --> 01:26:33.000
will get mad at them this did not happen with the internet it was not present and part of the thing is social media was not like it is today the kind of uh was it decentralization of media in general has caused this as well and also the fact of day trading there's so many different things that are different it's crazy i do think ai is different from the internet in part if you just measured it on the speed of adoption

1484
01:26:33.580 --> 01:26:35.161
Especially if we just think about generative AI.

1485
01:26:35.201 --> 01:26:36.321
I know AI is lots of things.

1486
01:26:36.381 --> 01:26:40.422
But the adoption of the internet required physical connections to your house.

1487
01:26:40.482 --> 01:26:43.323
The adoption of generative AI involves having a web browser.

1488
01:26:43.723 --> 01:26:46.544
It took a vast amount of effort to bring internet to people.

1489
01:26:46.804 --> 01:26:49.225
Even with dial-up connections, it still required the distribution.

1490
01:26:49.245 --> 01:26:53.346
And that's why it was so slow and there was less hype than AI.

1491
01:26:53.406 --> 01:26:54.927
I do agree that there's way more hype.

1492
01:26:55.127 --> 01:27:01.316
And we are, again, going back to this point that we're clustering AI in this big category of lots of different things.

1493
01:27:01.376 --> 01:27:02.337
There's generative AI.

1494
01:27:02.457 --> 01:27:03.258
There's generative AI.

1495
01:27:03.278 --> 01:27:04.400
There's like real world AI.

1496
01:27:04.420 --> 01:27:06.823
The generative AI is explicitly what I'm talking about here.

1497
01:27:06.843 --> 01:27:11.089
When bosses are saying you need to use AI, they're not saying I need you to go and buy a Unibeam robot.

1498
01:27:11.149 --> 01:27:14.613
They're saying use LLMs so that I – and that's the thing.

1499
01:27:14.733 --> 01:27:18.737
They have this theory, the era of the business idiot, where it's like we are ruled by people that don't do work.

1500
01:27:19.118 --> 01:27:26.386
Because nobody who actually does a bunch of work, who really is productive, is harassing someone who works for them for not being productive enough.

1501
01:27:26.866 --> 01:27:28.407
They don't have the time.

1502
01:27:28.427 --> 01:27:29.088
They're doing work.

1503
01:27:29.308 --> 01:27:35.793
Someone who is sitting there with the ingratiation machine that's telling them every beautiful idea out of their messy little skull is amazing.

1504
01:27:36.374 --> 01:27:38.215
Yeah, they're going, damn, this thing says I'm a genius.

1505
01:27:38.295 --> 01:27:40.797
Why are you not using the genius machine to do more work?

1506
01:27:41.438 --> 01:27:46.402
And yeah, if you're a boss that goes to lunch, leaves lunch, and sometimes reads your emails, LLMs are magic.

1507
01:27:46.982 --> 01:27:51.826
Do you know one of the most compelling arguments I have for the overhype of AI?

1508
01:27:52.526 --> 01:27:58.971
In a world where everybody has access to these tools, whatever the tools can do would largely be commoditized.

1509
01:27:59.112 --> 01:28:06.878
And what the tools can't do, which one could say is the human taste judgment, you could say it's people skills, whatever you want to say, is...

1510
01:28:07.578 --> 01:28:16.265
is now going to be the valuable thing because the scarce and the hard becomes the most valuable through history and the commoditized becomes the least valuable.

1511
01:28:16.585 --> 01:28:24.631
So the very nature that we're commoditizing the generation of content or whatever you want to call it code means that's actually not where the value will accrue for the user.

1512
01:28:25.352 --> 01:28:32.958
And actually, if you think about what it takes to now make something that is objectively great, if an AI can do it,

1513
01:28:34.453 --> 01:28:36.714
then the great thing is not of value.

1514
01:28:37.675 --> 01:28:47.361
So I think a lot, I've been thinking a lot actually about how do you avoid the temptation of slopification of the things you make, the value you put into the world.

1515
01:28:47.861 --> 01:28:50.463
It's a very simple example that people will be able to relate to.

1516
01:28:50.883 --> 01:28:57.367
If you use ChatGPT or Anthropic, you know, Claude, to make your LinkedIn posts, let's say, they will be shit LinkedIn posts.

1517
01:28:57.707 --> 01:28:58.948
Because everybody else is using them.

1518
01:28:59.048 --> 01:29:04.874
And actually, a great LinkedIn post now is someone who doesn't use them and makes something that's irreplaceably human.

1519
01:29:05.235 --> 01:29:05.535
Right.

1520
01:29:05.735 --> 01:29:11.541
And deeper and more personal, N of one, lived experience.

1521
01:29:11.861 --> 01:29:12.182
Yeah.

1522
01:29:12.282 --> 01:29:13.683
All these things that AI can't do.

1523
01:29:13.863 --> 01:29:15.705
And I think that's a compelling argument that actually...

1524
01:29:16.526 --> 01:29:19.187
The commodity tools produce commodity outcomes.

1525
01:29:19.207 --> 01:29:20.747
So everyone has access to these things.

1526
01:29:20.787 --> 01:29:21.447
And what's changed?

1527
01:29:21.587 --> 01:29:22.347
Like, really, like what?

1528
01:29:22.547 --> 01:29:23.808
The slopification of stuff.

1529
01:29:23.988 --> 01:29:25.008
We've got a bunch of slob.

1530
01:29:25.048 --> 01:29:27.168
But these people were half-arsing their jobs before.

1531
01:29:27.408 --> 01:29:28.909
It's just a half-arsery machine.

1532
01:29:29.529 --> 01:29:30.849
And it's just, it's the thing.

1533
01:29:31.029 --> 01:29:32.570
It's what I'm talking about with the slop blogs.

1534
01:29:32.910 --> 01:29:33.630
It's like, it's it.

1535
01:29:33.690 --> 01:29:38.171
Yeah, people that gave you dog shit before have now got the dog shit machine to pump out dog shit.

1536
01:29:38.491 --> 01:29:41.371
It's, so there's a guy called Carl Brown, Internet Bucks.

1537
01:29:41.652 --> 01:29:42.152
Awesome guy.

1538
01:29:42.512 --> 01:29:43.472
Great software engineer.

1539
01:29:43.972 --> 01:29:47.674
He said, I might have said this earlier, so it makes the easier things easier, the hard things harder.

1540
01:29:47.694 --> 01:29:52.037
When you know you're doing a really distinct, small script for something and it can plop that out, it's awesome.

1541
01:29:52.457 --> 01:29:54.538
I used Claude the other day for something useful.

1542
01:29:54.799 --> 01:29:55.979
My kid loves Minecraft.

1543
01:29:56.199 --> 01:29:59.281
I was trying to fix a fucking broken mod because he loves his Witherstorm.

1544
01:29:59.401 --> 01:29:59.962
It's awesome.

1545
01:30:00.782 --> 01:30:03.386
And it still took me half an hour and kept getting things wrong.

1546
01:30:03.606 --> 01:30:04.567
What do you use AI for?

1547
01:30:04.827 --> 01:30:05.308
Generative AI.

1548
01:30:05.348 --> 01:30:06.269
I really don't.

1549
01:30:06.389 --> 01:30:06.930
You don't use it?

1550
01:30:07.010 --> 01:30:13.158
With Bloomberg Terminal, I use Ask B, which is just when it's like requesting the consensus analyst estimates for NVIDIA.

1551
01:30:13.178 --> 01:30:14.099
But otherwise you don't use it?

1552
01:30:14.339 --> 01:30:14.519
No.

1553
01:30:14.860 --> 01:30:15.701
So how do you know it's bad?

1554
01:30:16.121 --> 01:30:16.682
I've used it.

1555
01:30:16.702 --> 01:30:18.083
I've put it through its paces.

1556
01:30:18.123 --> 01:30:21.606
I've used it to try and do financial models and found one error and immediately be like, ah.

1557
01:30:22.086 --> 01:30:23.888
I've never been particularly impressed.

1558
01:30:24.068 --> 01:30:27.371
The one thing I will defend it on is it's really good for tech support.

1559
01:30:27.691 --> 01:30:31.155
I had this thing called Synergy in my New York place I go to.

1560
01:30:31.395 --> 01:30:33.937
I had this monitor where I have a MacBook and a PC laptop.

1561
01:30:34.137 --> 01:30:39.442
And this thing, Synergy, for using the same mouse and keyboard, drop in a giant fucking...

1562
01:30:39.822 --> 01:30:57.317
troubleshooting log into this thing going what's wrong and it going this is wrong yeah super useful is that trillion dollars no is that a two trillion dollar company no better than google though right better than google no i mean uh yeah remember do you use google search still i try i have to fucking push the crap out of the way and oh my god my

1563
01:30:57.457 --> 01:31:00.560
I can't remember the last time I did a Google search.

1564
01:31:00.760 --> 01:31:02.622
Christ, I find myself using Bing sometimes.

1565
01:31:03.103 --> 01:31:03.463
I know.

1566
01:31:03.623 --> 01:31:04.524
I hate saying it too.

1567
01:31:04.965 --> 01:31:08.408
But I have to scroll past the AI crap because I want the good stuff.

1568
01:31:09.069 --> 01:31:12.612
I want the actual links to stuff so that I can read the thing and go.

1569
01:31:12.732 --> 01:31:15.215
But you can ask the AI to give you the links.

1570
01:31:15.475 --> 01:31:17.977
Yeah, and it doesn't do a particularly good job.

1571
01:31:17.997 --> 01:31:18.417
Like my...

1572
01:31:18.858 --> 01:31:22.981
So say that the other day, my iPad wasn't turning on and it was doing this funny little thing on the screen.

1573
01:31:23.361 --> 01:31:25.743
You think that it's better to type that into Google?

1574
01:31:25.823 --> 01:31:27.124
Oh, no, I must be clear.

1575
01:31:27.364 --> 01:31:29.966
That may be the only LLM use case I defend.

1576
01:31:30.166 --> 01:31:32.007
The troubleshooting thing is awesome for it.

1577
01:31:32.127 --> 01:31:33.748
It's the one weakness I have.

1578
01:31:33.808 --> 01:31:36.490
It's like genuinely being able to drop a log into it.

1579
01:31:36.831 --> 01:31:37.431
That's awesome.

1580
01:31:37.711 --> 01:31:40.133
Again, that is not what they're selling it as.

1581
01:31:40.273 --> 01:31:42.235
They're not selling it as a useful little tool.

1582
01:31:42.435 --> 01:31:43.476
They're selling it as the...

1583
01:31:44.636 --> 01:32:11.031
software as the thing that will change everything that will replace all jobs that will do this and that it's not like they sold it as a quirky bit of software no you are right they are you know telling us that it is going to replace everything but funnily enough the critics are saying that as well which one i mean i mean they are like the jeffrey hintons of the world you know even people that have left the safety team in chat gbt who've who've sat here with these are critics that are that are warning of the impacts it's going to have on the world

1584
01:32:11.291 --> 01:32:15.495
It's weird how all these critics also have a vested interest in AI doing well, though.

1585
01:32:15.555 --> 01:32:23.903
Daniel Former, open AI guy, AI 2027, written with the Star Codex guy that was nothing more than badly written science fiction that he's already had to walk back.

1586
01:32:23.923 --> 01:32:26.186
But you know he could have made more money by staying at Chachapi too.

1587
01:32:27.204 --> 01:32:49.630
could he i mean looks like if he had options early it sticking around did he lose the options how much did he lose you're not saying that they're they're being critical they're not critical of the companies themselves they're not critical of the stealing they're not critical of the environmental damage they're not critical of the fact that you cannot rely on the answers they're critical of this big scary boogeyman out in the future

1588
01:32:50.450 --> 01:32:56.273
where it's like, oh, I'm scared of when this becomes so powerful and everyone should talk to me about how scary and powerful it is.

1589
01:32:56.533 --> 01:32:58.434
They're not saying, hey, here are the harms today.

1590
01:32:58.735 --> 01:33:00.416
Here are the things we're actually looking at today.

1591
01:33:00.556 --> 01:33:15.324
Here are the social problems of having this automated way of spewing out slop, of filling our feeds with crap, of having information that will pop up that is presented, even with the little disclaimer thing of saying, yeah, sometimes this gets shit wrong in the tiniest...

1592
01:33:15.824 --> 01:33:16.424
words possible.

1593
01:33:16.744 --> 01:33:21.507
They don't talk about the fact that these things are trained on stealing millions of people's work.

1594
01:33:21.527 --> 01:33:26.369
But on that last point where you say that it's going to get progressively more intelligent and when it does, it will be a danger.

1595
01:33:26.389 --> 01:33:26.729
Yeah.

1596
01:33:27.549 --> 01:33:32.592
Would you agree with the statement that artificial intelligence has gotten more intelligent

1597
01:33:33.612 --> 01:33:37.456
If you measure it based on any sort of measure of intelligence one might use.

1598
01:33:37.597 --> 01:33:39.759
It's got better on the tests that are rigged for the models.

1599
01:33:39.999 --> 01:33:42.622
It's got better at tests where you can train for the test.

1600
01:33:42.963 --> 01:33:43.703
Okay, so it's got better at...

1601
01:33:44.224 --> 01:33:47.027
It's got better at tests that they're intentionally trained for.

1602
01:33:47.367 --> 01:33:51.932
So if you logged the rate of improvement on a graph, it would look something like...

1603
01:33:52.834 --> 01:33:53.074
this.

1604
01:33:53.615 --> 01:33:53.995
Right.

1605
01:33:54.036 --> 01:33:54.336
You agree?

1606
01:33:54.356 --> 01:33:56.338
In terms of what it's capable of doing.

1607
01:33:56.759 --> 01:33:58.040
Oh, there we go.

1608
01:33:58.160 --> 01:33:58.341
Yeah.

1609
01:33:58.481 --> 01:34:00.864
Because it's not got new features.

1610
01:34:00.884 --> 01:34:09.134
You'll notice that outside of OpenAI and Anthropic, when you remove the coding startups, there's basically no successful AI startup company.

1611
01:34:09.354 --> 01:34:12.558
So we agree that it's got better, it's got more capable

1612
01:34:14.002 --> 01:34:14.883
At doing things.

1613
01:34:15.284 --> 01:34:16.586
Yeah, fine.

1614
01:34:16.606 --> 01:34:17.807
Over time, AI's got more capable.

1615
01:34:18.248 --> 01:34:29.382
If we imagine that trajectory will continue, it will get more capable, then at some point it does cross, you know, this is what they say to me, it crosses human intelligence.

1616
01:34:29.422 --> 01:34:30.223
And at such time...

1617
01:34:31.194 --> 01:34:35.078
Will it not start to do some of the jobs that people are doing today?

1618
01:34:35.398 --> 01:34:36.520
Outside of software engineering.

1619
01:34:36.540 --> 01:34:37.120
Remove software.

1620
01:34:37.140 --> 01:34:38.682
Because I will concede software engineering.

1621
01:34:38.702 --> 01:34:39.563
It's got better at that.

1622
01:34:39.863 --> 01:34:40.984
Outside of software engineering.

1623
01:34:41.004 --> 01:34:41.124
Yeah.

1624
01:34:41.345 --> 01:34:41.625
Where?

1625
01:34:42.025 --> 01:34:43.687
So the chief of staff things, the admin.

1626
01:34:44.067 --> 01:34:44.308
Okay.

1627
01:34:44.348 --> 01:34:45.509
So it's got better at admin.

1628
01:34:45.569 --> 01:34:47.631
Video generation, photo generation.

1629
01:34:47.651 --> 01:34:47.971
Okay.

1630
01:34:48.212 --> 01:34:50.014
Text generation, theoretically.

1631
01:34:50.054 --> 01:34:50.214
Coding.

1632
01:34:50.574 --> 01:34:50.955
Coding.

1633
01:34:51.615 --> 01:34:51.835
Right.

1634
01:34:52.275 --> 01:34:54.076
And then I'd say agentic workflows.

1635
01:34:54.336 --> 01:34:55.656
What is an agentic workflow?

1636
01:34:55.716 --> 01:34:57.737
So automated workflows where you're doing the same.

1637
01:34:58.117 --> 01:35:01.778
I mean, a good example is looking at the backend data of the diary of a CEO.

1638
01:35:01.999 --> 01:35:02.599
Summarizing.

1639
01:35:03.019 --> 01:35:04.419
Looking at all of the data, ingesting all of it.

1640
01:35:04.559 --> 01:35:06.260
Going out onto the internet and searching who Ed is.

1641
01:35:06.740 --> 01:35:08.821
Looking at every interview you've ever done ever.

1642
01:35:08.841 --> 01:35:11.902
This is summarizing and generating.

1643
01:35:12.162 --> 01:35:14.903
Making a little model on the things people want to know from Ed.

1644
01:35:15.543 --> 01:35:17.744
Producing a report, sending that to my inbox.

1645
01:35:17.824 --> 01:35:20.885
Me getting a 20, 30, 40, 50 page report on Ed before he arrives.

1646
01:35:21.465 --> 01:35:23.746
This is all basically the same thing it's been doing for years, though.

1647
01:35:24.046 --> 01:35:25.527
It's not really new capabilities.

1648
01:35:25.547 --> 01:35:27.788
Research, it's search.

1649
01:35:27.948 --> 01:35:28.908
It's still the same thing.

1650
01:35:28.948 --> 01:35:30.029
They've had web search for years.

1651
01:35:30.109 --> 01:35:31.929
They've had report generation for years.

1652
01:35:32.169 --> 01:35:33.110
But we couldn't generate...

1653
01:35:34.667 --> 01:35:59.253
high quality videos that are like indistinguishable from cameras see dance and these ones that look like movies i mean they are incredible so i'm saying the point i'm trying to make is that if we imagine that over the last 10 years there has been a rate of improvement in terms of capabilities and output and quality we've seen hallucinations drop we've seen the models get more quote-unquote intelligent get better you know if you didn't give it an iq test it's getting higher scores than it was 10 years ago we agree that there's been

1654
01:35:59.713 --> 01:36:01.534
an upward motion of improvement.

1655
01:36:01.554 --> 01:36:03.935
This is pretty much how machine learning goes when you feed it more data.

1656
01:36:04.015 --> 01:36:04.415
Exactly.

1657
01:36:04.495 --> 01:36:05.595
And you put more compute behind it.

1658
01:36:06.175 --> 01:36:11.017
So if this continues, what does the future look like?

1659
01:36:11.417 --> 01:36:14.538
So the rebuttal I was expecting to hear is that it won't continue.

1660
01:36:14.798 --> 01:36:16.239
And I actually don't think it...

1661
01:36:16.379 --> 01:36:18.900
I think that there are hard limits that we're going to hit.

1662
01:36:18.920 --> 01:36:21.241
So you do believe that there's a hard limit somewhere?

1663
01:36:21.301 --> 01:36:24.522
We've kind of already hit the diminishing returns level because...

1664
01:36:25.162 --> 01:36:30.647
For example, video generation, which is, by the way, far less an American concern anymore.

1665
01:36:30.807 --> 01:36:32.389
OpenAI has shut down Sora.

1666
01:36:32.429 --> 01:36:33.710
I think you can still use the API.

1667
01:36:33.790 --> 01:36:37.954
But nevertheless, look around you with the amount of stuff and the crew you need to get a shot.

1668
01:36:38.234 --> 01:36:41.657
People think the movies are just shot by shot by shot, and they just magically happen.

1669
01:36:41.837 --> 01:36:43.659
When you've got a wonderful girlfriend of...

1670
01:36:43.919 --> 01:36:47.442
first ADs, assistant directors, you've got gaffers, you've got lighters.

1671
01:36:47.782 --> 01:36:50.524
And also, simulating light is insanely difficult.

1672
01:36:50.584 --> 01:36:57.389
There are so many magical things that happen in creating visual images that, yeah, you could create a one-minute-long thing that might fool someone.

1673
01:36:57.849 --> 01:36:59.630
How do you practically turn that into a movie?

1674
01:36:59.710 --> 01:37:02.572
Because that movie, I forget what the name is, there was a movie that claimed it...

1675
01:37:03.433 --> 01:37:26.380
aired at can it didn't no one it aired in the city of can during the can film festival it was not at the film festival when it comes to the practical creation of actual things at the end of it versus magic tricks the actual practical outcomes are not there the reason i keep coming back to the capabilities thing for the example is yeah they can do better at tests do better you number go up when it comes to can this actually do distinct tasks you can rely on it

1676
01:37:27.220 --> 01:37:28.460
You can rely on it for summaries.

1677
01:37:28.580 --> 01:37:29.981
You can rely on it for generations.

1678
01:37:30.041 --> 01:37:33.641
The things it was doing, it's getting linearly-ish better at.

1679
01:37:33.661 --> 01:37:36.122
But again, there's a ceiling to that.

1680
01:37:36.142 --> 01:37:38.362
Like, okay, so it gets really good at research.

1681
01:37:38.382 --> 01:37:39.263
What does that actually mean?

1682
01:37:39.603 --> 01:37:41.423
You've already kind of got the automation there.

1683
01:37:41.803 --> 01:37:42.943
What is the next step of that?

1684
01:37:42.983 --> 01:37:47.004
Because training it to be more autonomous, for example, that's not something that comes from training data.

1685
01:37:47.064 --> 01:37:50.665
That is actually a new, Gary Marcus, a neurosymbolic.

1686
01:37:50.685 --> 01:37:53.606
You actually need to build a structure around the AI to make it work.

1687
01:37:53.846 --> 01:37:54.826
And even then, it's...

1688
01:37:54.826 --> 01:37:55.626
It doesn't fix the...

1689
01:37:55.686 --> 01:38:01.510
So you're saying that there will become a point where the rate of improvement will plateau... We're already there.

1690
01:38:01.530 --> 01:38:02.070
...and stop.

1691
01:38:02.190 --> 01:38:05.312
We've already hit that diminishing... Gary Marcus said this in 2022 as well.

1692
01:38:05.612 --> 01:38:11.215
You know there's lots of people listening now that, like, they've had their workflows completely transformed by these tools.

1693
01:38:11.455 --> 01:38:11.815
Have they?

1694
01:38:12.235 --> 01:38:13.516
There'll be... Yeah, there are.

1695
01:38:13.656 --> 01:38:17.578
Yeah, the thing is, first of all, every single one of them, did you pay for the tokens?

1696
01:38:17.899 --> 01:38:18.439
That's the thing.

1697
01:38:18.499 --> 01:38:19.359
Did you pay for the tokens?

1698
01:38:19.399 --> 01:38:19.840
And also...

1699
01:38:20.660 --> 01:38:21.661
How many tokens did you burn?

1700
01:38:21.701 --> 01:38:23.381
But putting all that aside, what workflows?

1701
01:38:23.442 --> 01:38:25.803
Because of it, yeah, I did a bunch of web scraping or web searches.

1702
01:38:26.063 --> 01:38:27.063
I'm just not impressed.

1703
01:38:27.444 --> 01:38:28.964
Did you make an entire fucking movie?

1704
01:38:29.225 --> 01:38:29.725
No, you didn't.

1705
01:38:30.305 --> 01:38:31.686
Is it speeding up your coding?

1706
01:38:32.146 --> 01:38:32.907
Yeah, I believe that.

1707
01:38:33.167 --> 01:38:34.427
I've heard that from multiple people.

1708
01:38:34.587 --> 01:38:35.088
But again...

1709
01:38:35.990 --> 01:38:37.111
How much can you trust this?

1710
01:38:37.291 --> 01:38:49.463
I think what I was getting at is, you know, in the moment of any technological innovation, people, they extrapolate linearly or they view it as a static state, i.e.

1711
01:38:49.483 --> 01:38:53.427
they think today is going to look like tomorrow or they think it's going to get better in this sort of straight line.

1712
01:38:53.727 --> 01:38:56.510
But what we end up seeing a lot of the time is this exponential improvement.

1713
01:38:56.850 --> 01:39:02.655
All of the innovations we're talking about with you, with compute and all that, with fast processes, those are hardware breakthroughs.

1714
01:39:03.135 --> 01:39:07.038
The hardware breakthrough companies don't seem to be fixing the LLM problems.

1715
01:39:07.318 --> 01:39:09.800
Despite all the king's horses, all the king's men, we're what?

1716
01:39:10.220 --> 01:39:12.802
Nine, ten generations of TPUs from Google now.

1717
01:39:13.183 --> 01:39:16.005
Broadcoms building stuff with OpenAI, their jalapeno chip.

1718
01:39:16.565 --> 01:39:19.948
And yet none of these people can just say, yeah, we're on the path to making this profitable.

1719
01:39:20.428 --> 01:39:21.169
Because they can't.

1720
01:39:21.409 --> 01:39:26.553
If we fix the environmental problems and the profitability situation, maybe I'd be more generous with this stuff.

1721
01:39:27.093 --> 01:39:29.836
But they don't seem to be able to.

1722
01:39:29.936 --> 01:39:38.723
And you talk about these improvements in capabilities, there's a certain point at which I'm saying, okay, can it do even a tenth of the stuff they're promising?

1723
01:39:38.903 --> 01:39:44.968
Samuel, when the other fucking week was saying, it was going to be in like six months, we'll be like a genie that you can ask wishes for.

1724
01:39:45.168 --> 01:39:46.990
It's like, motherfuckers never watched Aladdin.

1725
01:39:47.010 --> 01:39:47.770
What's he talking about?

1726
01:39:47.810 --> 01:39:47.990
Like...

1727
01:39:48.711 --> 01:39:50.092
Also, the genie was charming.

1728
01:39:50.152 --> 01:39:56.354
Anyway, long story short, the promises do not line up with the capabilities or the capability improvements.

1729
01:39:56.434 --> 01:40:05.237
An exponential improvement in software and software performance is always a result of direct hardware improvement.

1730
01:40:05.617 --> 01:40:10.259
We have all the gifted mathematicians, all the gifted software engineers, all the gifted hardware engineers.

1731
01:40:10.979 --> 01:40:11.699
And where are we?

1732
01:40:12.199 --> 01:40:13.840
Trillion plus dollars in with...

1733
01:40:14.729 --> 01:40:18.273
The future great financial crisis and the world's greatest marketing psyop.

1734
01:40:18.433 --> 01:40:24.860
I just think in the future, I do think that all of the devices and the computers we use and the physical items in our world will be more intelligent.

1735
01:40:25.341 --> 01:40:27.643
I mean, sure, but is that LLMs?

1736
01:40:27.984 --> 01:40:30.506
That'll be powered by the underlying AI infrastructure.

1737
01:40:30.707 --> 01:40:33.089
It'll be the more data centers.

1738
01:40:33.149 --> 01:40:34.290
It'll be energy coming down.

1739
01:40:34.451 --> 01:40:34.931
How does a...

1740
01:40:35.912 --> 01:40:47.261
GPU full data center translate to a Nikon camera that can, I don't know even what you'd think, because what is the thing we're talking about here?

1741
01:40:47.281 --> 01:40:52.865
Because the idea that devices will get smarter, sure, I can see that it's a very broad statement.

1742
01:40:52.885 --> 01:40:53.586
I can see it happening.

1743
01:40:53.606 --> 01:40:54.486
It's really kind of happening.

1744
01:40:54.987 --> 01:40:56.608
What does that have to do with the data centers?

1745
01:40:56.628 --> 01:40:57.669
Because these data centers, again,

1746
01:40:58.009 --> 01:41:01.494
are not being built to make your consumer electronics smarter.

1747
01:41:01.734 --> 01:41:07.601
They're not being built for anything other than speculating on the ability to capture demand for generative AI services.

1748
01:41:07.782 --> 01:41:08.783
But it's not just generative AI.

1749
01:41:08.803 --> 01:41:09.344
We went through that.

1750
01:41:09.364 --> 01:41:09.604
It is.

1751
01:41:09.644 --> 01:41:12.448
No, but those data centers, they are being built for generative AI.

1752
01:41:12.608 --> 01:41:14.150
They are not being built for anything else.

1753
01:41:14.686 --> 01:41:34.658
Would you consider generative AI to be the fact that on Meta's earnings call, like a couple of weeks ago, Mark Zuckerberg said the big breakthrough we've had, which has resulted in 15 basis points of increased retention, I believe he was referring to Instagram, is that we now take anything you post on social media and we run it through an AI to get full context of what it is.

1754
01:41:34.838 --> 01:41:38.520
And because we can see a guy sat in front of me called Ed with...

1755
01:41:39.020 --> 01:42:07.200
blue shirt and coffee we now can train the ai to serve whoever wants blue shirt adding with coffee to the right user which means people are retained longer because isn't 15 basis points like 0.15 yeah it's not cool but it makes a difference at scale it makes a big difference at scale yeah but a hundred and something billion dollars in and the best you've got is 0.15 if he could be fired i mean how much of a difference because there's a reason he's saying basis points versus dollars

1756
01:42:08.000 --> 01:42:09.320
Because think about it like this.

1757
01:42:09.900 --> 01:42:10.921
If Mark Zuckerberg was...

1758
01:42:11.521 --> 01:42:12.561
I take your point about scale.

1759
01:42:13.041 --> 01:42:26.824
I'm saying the point I was making was that that is another application of these data centers because it needs a data center that is driving revenues, but also that's outside of us thinking about just generating a sentence.

1760
01:42:26.844 --> 01:42:29.285
And that's the gem, the generative model.

1761
01:42:29.645 --> 01:42:30.445
Muse.

1762
01:42:30.665 --> 01:42:32.205
Oh, Muse Spark is their LLM.

1763
01:42:32.285 --> 01:42:33.285
Gem is their generative...

1764
01:42:33.466 --> 01:42:34.126
It's Muse.

1765
01:42:34.386 --> 01:42:35.006
Well, Muse...

1766
01:42:35.726 --> 01:42:40.909
then that's them doing the weird thing where it's like on Instagram and it's like, Dave the cat, why is Dave the cat suffering?

1767
01:42:41.149 --> 01:42:42.629
Like it's the weird pop-up things.

1768
01:42:43.210 --> 01:42:44.010
Meta is fuck.

1769
01:42:44.230 --> 01:42:45.371
God damn that company sucks.

1770
01:42:45.391 --> 01:42:47.392
Like every time I think about how they've ruined that product.

1771
01:42:47.432 --> 01:42:48.452
But that's the thing though.

1772
01:42:48.592 --> 01:42:51.734
Again, why can't he just say with his whole chest, we've made a couple billion?

1773
01:42:52.274 --> 01:42:53.035
Why can't he say that?

1774
01:42:53.175 --> 01:42:53.795
Because he isn't.

1775
01:42:53.895 --> 01:42:56.897
Because there's not actually a way of going, I spent all this money.

1776
01:42:57.258 --> 01:43:02.762
I spent $14 billion goddamn dollars on scale by Alexander Wong, and I made this much.

1777
01:43:02.862 --> 01:43:03.322
They can't...

1778
01:43:03.863 --> 01:43:17.553
It gets back to a very simple point of, hey, if it was going well, you'd tell me how well it was going, rather than, I don't know, doing this weird rain dance thing where you're like, well, if we move all the pieces around in three years, theoretically, this will happen.

1779
01:43:17.573 --> 01:43:17.813
Yeah.

1780
01:43:19.322 --> 01:43:23.404
I've done almost 700 interviews with some of the most interesting people in the world.

1781
01:43:23.484 --> 01:43:29.147
And one of the things you learn, which is unexpected, is that vulnerability is the doorway to connection.

1782
01:43:29.167 --> 01:43:34.270
And after sitting here for two, three hours with a guest, I feel a deep sense of connection to them.

1783
01:43:34.390 --> 01:43:40.914
And as they leave, what I get them to do is to write a question in the diary of a CEO.

1784
01:43:40.994 --> 01:43:43.615
We've taken all of the questions from the diary of a CEO.

1785
01:43:43.756 --> 01:43:45.697
We have put the question...

1786
01:43:46.557 --> 01:43:49.778
here on this card with the name of the person that wrote it.

1787
01:43:49.918 --> 01:43:54.660
So you can sit at home as I do with my fiancee and my colleagues at work and other people in my life.

1788
01:43:55.000 --> 01:44:01.922
Whenever we get a minute, we play the Diary of a CEO conversation cards and it is incredible what happens.

1789
01:44:02.002 --> 01:44:05.563
These are great if you're in a romantic relationship and you wanna connect your partner

1790
01:44:05.743 --> 01:44:09.286
These are also great if you're in a team and you want to bond your team together.

1791
01:44:09.526 --> 01:44:20.574
And I have to say they're also great for families that want to learn more about each other and that need a good excuse to spend some time in a digital world, in the analog environment, connecting human to human.

1792
01:44:20.834 --> 01:44:25.138
It is remarkable what the right question at the right time can do.

1793
01:44:25.318 --> 01:44:29.461
Go to thediary.com and you can get these conversation cards

1794
01:44:31.634 --> 01:44:36.358
I do think you're accurate and right when you talk about the fact that there's a lot of like, is the word for gazy?

1795
01:44:36.679 --> 01:44:36.879
Yeah.

1796
01:44:37.319 --> 01:44:41.002
Where like there's a lot of people that have spent a lot of money and they kind of shouldn't have spent it and they fucked up.

1797
01:44:41.383 --> 01:44:43.965
And now they're thinking, shit, like we've spent all this invested money.

1798
01:44:43.985 --> 01:44:45.887
Kind of like the metaverse was a bit of a joke.

1799
01:44:46.007 --> 01:44:46.487
Oh my God.

1800
01:44:46.507 --> 01:44:47.068
That was a bit of a joke.

1801
01:44:47.088 --> 01:44:47.808
That's so weird.

1802
01:44:47.888 --> 01:44:48.709
A lot of money spent.

1803
01:44:48.849 --> 01:44:51.071
We kind of thought this dream was coming of this one.

1804
01:44:51.111 --> 01:44:52.853
I shouldn't say dream because it's not a dream that I've had.

1805
01:44:53.473 --> 01:44:54.094
A dream that they had.

1806
01:44:54.394 --> 01:44:57.017
Yeah, this sort of virtual world, and actually it never transpired.

1807
01:44:57.057 --> 01:44:59.080
And there's no sign that it will in the near term.

1808
01:44:59.660 --> 01:45:03.485
AI and the dot-com boom, in this regard, are the same.

1809
01:45:03.765 --> 01:45:04.766
NFTs were the same.

1810
01:45:05.668 --> 01:45:05.928
Right.

1811
01:45:06.028 --> 01:45:08.891
So one could argue that a lot of the crypto industry was the same.

1812
01:45:09.212 --> 01:45:10.974
It's Wang that is inflated by the media.

1813
01:45:11.334 --> 01:45:13.997
The difference is, the reason the metaverse and NFTs didn't...

1814
01:45:14.658 --> 01:45:17.159
escape this was there weren't stocks to speculate on.

1815
01:45:17.480 --> 01:45:19.641
There weren't big companies that you could invest in.

1816
01:45:19.921 --> 01:45:21.862
They had record earnings in 2021.

1817
01:45:21.902 --> 01:45:29.706
There's a bunch of money floating in the system thanks to post-COVID, the PRDC, that basically government, federal money flowed in to the banks.

1818
01:45:29.726 --> 01:45:32.127
There's a bunch of easy money, zero interest for a year.

1819
01:45:32.467 --> 01:45:33.268
Money was easy to find.

1820
01:45:33.448 --> 01:45:34.748
Then after that, there was the hangover.

1821
01:45:34.928 --> 01:45:36.789
Growth started to slow down dramatically.

1822
01:45:36.989 --> 01:45:41.472
This is actually my rock-com bubble theory, which is they don't have any hyper-growth ideas anymore.

1823
01:45:42.052 --> 01:45:44.597
So suddenly they started buying GPUs.

1824
01:45:44.758 --> 01:45:46.241
And when they bought GPUs, people went...

1825
01:45:46.752 --> 01:45:47.553
They're doing AI.

1826
01:45:48.293 --> 01:45:49.334
Oh, we better buy the stock.

1827
01:45:49.374 --> 01:45:50.595
And the stock's won an incredible run.

1828
01:45:50.835 --> 01:45:52.296
Meta's like several hundred percent.

1829
01:45:52.817 --> 01:45:56.900
In the last few years, their stock has grown by hundreds of percent, despite zero proof.

1830
01:45:57.500 --> 01:46:03.225
And because the media was just saying, yeah, Meta's revenue is growing because of AI, right?

1831
01:46:03.325 --> 01:46:05.266
Microsoft's revenue is growing because of AI, right?

1832
01:46:05.687 --> 01:46:09.590
The fugazi you're talking about was the fact that everyone just gave them credit in advance.

1833
01:46:09.890 --> 01:46:11.891
And now we're kind of getting to the point where it's like, hey...

1834
01:46:12.792 --> 01:46:15.594
You didn't spend that trillion dollars for no reason, did you, Satya?

1835
01:46:16.195 --> 01:46:20.038
Amy Hood, just going to take him out back, send him to the glue factory or something?

1836
01:46:20.358 --> 01:46:21.839
I do think there's overspending.

1837
01:46:22.159 --> 01:46:23.000
I want to concede that.

1838
01:46:23.040 --> 01:46:23.700
Dramatic.

1839
01:46:24.021 --> 01:46:25.121
Yeah, no, I do think there is.

1840
01:46:25.242 --> 01:46:30.245
And I think the reason why there's overspending, Ed, is I think there is something here.

1841
01:46:31.246 --> 01:46:36.070
And in terms of like, I think there is practical uses for this technology.

1842
01:46:36.390 --> 01:46:42.614
And I think when people realize that through history, they go crazy because they want to be the person that owns the opportunity.

1843
01:46:42.634 --> 01:46:43.455
I'm going to be honest.

1844
01:46:43.515 --> 01:46:45.136
I just, I fundamentally don't agree.

1845
01:46:45.196 --> 01:46:46.217
You don't agree with which part.

1846
01:46:46.517 --> 01:46:49.980
I don't agree that the speculation is a result of actual demand.

1847
01:46:50.100 --> 01:46:51.181
I don't believe it's suspect.

1848
01:46:51.462 --> 01:46:55.605
I don't think private credit is sinking hundreds of billions of dollars into AI because of actual demand.

1849
01:46:55.825 --> 01:47:02.852
They are doing it because they saw the biggest companies in the world building data centers, making a ton of money from two companies they feed money, and went, I want some of that money.

1850
01:47:03.152 --> 01:47:06.775
I'm saying that I do think there is value in the underlying technology.

1851
01:47:06.935 --> 01:47:07.316
Sure.

1852
01:47:07.716 --> 01:47:08.357
And so I think...

1853
01:47:08.997 --> 01:47:10.078
I'm not saying how much value.

1854
01:47:10.278 --> 01:47:10.518
Right.

1855
01:47:10.638 --> 01:47:11.479
Okay.

1856
01:47:11.840 --> 01:47:12.660
I get you meaning that.

1857
01:47:12.680 --> 01:47:13.121
That's fair.

1858
01:47:13.141 --> 01:47:15.062
I'm not saying it's proportionate to the investment.

1859
01:47:15.603 --> 01:47:17.504
All I'm saying is that, do you know what it's like?

1860
01:47:17.705 --> 01:47:20.647
It's like, if I take your example, the Rot Economy essay that you wrote.

1861
01:47:20.667 --> 01:47:20.847
Yeah.

1862
01:47:21.188 --> 01:47:24.991
Say that you're on a desert island and then someone says they found a banana tree.

1863
01:47:25.231 --> 01:47:25.471
Right.

1864
01:47:26.232 --> 01:47:28.894
And there's 10,000 people on the island.

1865
01:47:28.914 --> 01:47:29.154
Okay.

1866
01:47:30.115 --> 01:47:34.740
They are going to stam fucking peed towards where they think the banana tree is.

1867
01:47:34.780 --> 01:47:37.303
They are going to fucking claw each other to pieces.

1868
01:47:37.543 --> 01:47:43.590
And if your essay here is right, that there was desperation because they hadn't found an innovation in a while, maybe that explains it.

1869
01:47:43.630 --> 01:47:44.931
Maybe there is a bit of value here.

1870
01:47:44.951 --> 01:47:45.392
Right.

1871
01:47:45.672 --> 01:47:51.018
And they're stam fucking peeding and killing each other and making irrational decisions like hungry people would.

1872
01:47:51.238 --> 01:47:53.281
I actually think then we actually agree.

1873
01:47:53.341 --> 01:47:58.907
That is actually my point, which is these three companies in matter, their main business lines are running out of growth.

1874
01:47:58.927 --> 01:48:00.048
There's only so much they can grow.

1875
01:48:00.068 --> 01:48:07.878
And indeed, in the next three and a half years, analysts think that these two bastards, these two, OpenAI and Anthropic, are going to spend over $400 billion on these people alone.

1876
01:48:08.378 --> 01:48:08.979
Microsoft...

1877
01:48:09.079 --> 01:48:10.080
Google and Amazon.

1878
01:48:10.100 --> 01:48:12.903
And the crazy thing is, is that's a large part of their future growth.

1879
01:48:12.963 --> 01:48:15.806
And if this money isn't spent, their growth slows down.

1880
01:48:16.027 --> 01:48:16.227
Okay.

1881
01:48:16.327 --> 01:48:18.149
So your point about a bananas, I actually agree.

1882
01:48:18.189 --> 01:48:19.150
That is the rock-on bubble.

1883
01:48:19.190 --> 01:48:21.252
It's they don't have a new thing and they're desperate.

1884
01:48:21.673 --> 01:48:24.476
And indeed, they got rewarded for buying the GPUs.

1885
01:48:24.756 --> 01:48:28.000
They got, when they bought these goddamn GPUs from NVIDIA,

1886
01:48:28.460 --> 01:48:30.661
All the markets went rock hard overnight.

1887
01:48:30.681 --> 01:48:31.301
They loved it.

1888
01:48:31.501 --> 01:48:37.063
There were stories about how they were sending armored cars with the GPUs to Microsoft to make sure Microsoft got the GPUs.

1889
01:48:37.463 --> 01:48:41.464
And so everyone saw all that money flowing in, even though they never disclosed AI revenues.

1890
01:48:41.664 --> 01:48:43.105
They saw the expenditures and they went...

1891
01:48:43.905 --> 01:48:45.746
Well, I want to do what these people are doing.

1892
01:48:45.767 --> 01:48:47.528
I want to get a little of that money, don't I?

1893
01:48:47.908 --> 01:48:56.995
I think the area where we have a slight disagreement is that I think the underlying technology has a lot more promise over the long term than you do.

1894
01:48:57.315 --> 01:48:59.097
So the thing I want to push back on there is...

1895
01:49:00.436 --> 01:49:12.321
To have progress with AI, just taking it in a vacuum, to have progress for these two companies to keep going and to keep progressing, they need to spend tens of billions of dollars a year on training.

1896
01:49:12.961 --> 01:49:20.264
The only way that that can happen is if these companies and venture capitalists and private credit firms and NVIDIA keep circulating money to them.

1897
01:49:20.504 --> 01:49:21.585
So the progress...

1898
01:49:22.285 --> 01:49:25.727
that we've got so far is entirely a result of this circular system.

1899
01:49:26.127 --> 01:49:29.149
So it means that without- Circular, you talked about VCs there.

1900
01:49:29.389 --> 01:49:31.491
Venture capitalists who are, by the way- Investors.

1901
01:49:31.651 --> 01:49:38.055
The majority of the funding that OpenAI got in the last six months came from SoftBank, NVIDIA, and Amazon.

1902
01:49:38.475 --> 01:49:39.015
Okay, yeah.

1903
01:49:39.156 --> 01:49:42.238
So just the point is, you're talking about progress continuing.

1904
01:49:42.738 --> 01:49:47.221
Progress in LLMs can only continue as long as the money keeps flowing.

1905
01:49:47.521 --> 01:49:48.502
I completely agree.

1906
01:49:48.822 --> 01:49:50.923
Once the money stops flowing, the progress stops.

1907
01:49:50.963 --> 01:49:54.246
But isn't that most like early... Spotify didn't make money for 20 years.

1908
01:49:54.346 --> 01:49:56.947
Spotify didn't lose $20.9 billion in one year.

1909
01:49:57.168 --> 01:50:00.770
They didn't need to raise $217 billion in the space of six months.

1910
01:50:00.970 --> 01:50:02.391
Yeah, and Uber's another example.

1911
01:50:02.411 --> 01:50:05.932
$33 billion since inception before it became a messy kind of profitable.

1912
01:50:05.992 --> 01:50:11.155
Amazon Web Services between 2003 and 2015, when it became profitable, $29.7 billion.

1913
01:50:11.315 --> 01:50:12.375
They'd spent.

1914
01:50:12.415 --> 01:50:14.096
Yeah, that's the total capital expenditures.

1915
01:50:14.136 --> 01:50:15.937
And that's not just Amazon Web Services.

1916
01:50:15.977 --> 01:50:17.518
That's the entire logistics operation.

1917
01:50:17.918 --> 01:50:18.878
Normalized for inflation.

1918
01:50:19.198 --> 01:50:21.820
So they all lost money for a long period of time, is the TLDR.

1919
01:50:21.860 --> 01:50:24.641
Yes, but the amount of money they lost is...

1920
01:50:26.215 --> 01:50:39.384
completely just magnitudes different on a level where these three come... Can I argue then that that's because the potential of intelligence permeates everything, whereas Amazon at the time was like selling books.

1921
01:50:39.944 --> 01:50:40.584
No.

1922
01:50:40.624 --> 01:50:42.065
That was bringing retail online.

1923
01:50:42.145 --> 01:50:44.867
When Amazon Web Services grew, it was... Oh, Amazon Web Services.

1924
01:50:44.887 --> 01:50:45.788
Amazon Web Services.

1925
01:50:45.928 --> 01:50:46.468
Amazon Cloud.

1926
01:50:46.668 --> 01:50:50.891
With Amazon Web Services, the reason I bring that up, I'm going to repeat something, but it's really important.

1927
01:50:50.911 --> 01:50:51.612
2003, it was founded.

1928
01:50:51.732 --> 01:50:51.992
Mm-hmm.

1929
01:50:52.132 --> 01:50:57.756
And it was founded mostly because Amazon, as a growing online store, needed hardcore infrastructure.

1930
01:50:57.796 --> 01:51:00.398
2006, I think, is when they turned it client-facing.

1931
01:51:00.518 --> 01:51:01.599
I may be wrong on the dates there.

1932
01:51:01.699 --> 01:51:03.380
But 2015 was the year it became profitable.

1933
01:51:03.981 --> 01:51:10.566
The total capital expenditures normalized for inflation were $29.7 billion across that 12-year period.

1934
01:51:11.286 --> 01:51:12.547
And yeah, it lost money.

1935
01:51:12.587 --> 01:51:15.549
But if we speak cold economics here...

1936
01:51:16.490 --> 01:51:18.693
Amazon didn't have to go into the...

1937
01:51:18.713 --> 01:51:24.120
They were unprofitable in a way, but their margins actually started improving because AWS was a very margin-heavy business.

1938
01:51:24.140 --> 01:51:26.142
It was great.

1939
01:51:26.443 --> 01:51:28.926
These two, Google, cash flow negative.

1940
01:51:29.166 --> 01:51:30.227
Amazon, cash flow negative.

1941
01:51:30.388 --> 01:51:34.733
These businesses, the reason you liked software businesses was they are meant to be...

1942
01:51:35.669 --> 01:51:37.270
Cash-heavy asset light.

1943
01:51:37.791 --> 01:51:48.479
These companies, along with Meta, have added more than $700 billion of new property, plants, and equipment, so assets, data centers, GPUs, in the last four years.

1944
01:51:48.799 --> 01:51:52.723
They have gone from being these cash machines to these cash...

1945
01:51:53.323 --> 01:52:20.067
furnaces you said a second ago this can only continue if investors continue to invest yes and i was saying i think that investors are used to pumping money into things that are burning cash your rebuttal to me sounds like well this is burning more cash than ever and then so i would say well is the opportunity bigger than those other case studies you reference like aws and one would say that the opportunity of intelligence is

1946
01:52:21.242 --> 01:52:40.706
permeates everything so the TAM the total addressable market is enormous maybe the rebuttal back to me is about open source and all these kinds no no no I actually know what you're going at so what you were describing there is the argument that Satya Nadella or Sam Alton would make the theoretical opportunity of large language models and I could have bought that shit into any 24 from them

1947
01:52:41.226 --> 01:52:42.947
when they were like, oh, we see the opportunity.

1948
01:52:43.307 --> 01:52:48.851
We've gone way past the point at which you can rationally argue that LLMs need this much money.

1949
01:52:49.171 --> 01:52:54.974
And when I say the money needs to keep flowing, I am talking these two companies, OpenAI, just OpenAI.

1950
01:52:55.694 --> 01:53:05.460
Clammy Sammorton has said, Wall Street Journal and Issa Gardizi reported a few weeks ago, they plan to spend $750 billion on compute through 2030.

1951
01:53:05.500 --> 01:53:08.582
I think they're going to be dead before then, but $750 billion.

1952
01:53:10.643 --> 01:53:12.485
That is an insane amount of money.

1953
01:53:12.505 --> 01:53:13.085
It is crazy.

1954
01:53:13.205 --> 01:53:15.567
And a large chunk of that is training.

1955
01:53:15.627 --> 01:53:21.792
So when I say progress, I mean literally to make the models better at stuff requires billions of dollars invested just in data.

1956
01:53:23.051 --> 01:53:25.533
and also tens of billions of dollars of taking that data.

1957
01:53:25.913 --> 01:53:35.080
And so training, training is actually a really interesting thing, because when you think of like, for Jake and Troy, my trainers, when I train with them, when I lift with them, I have a defined thing.

1958
01:53:35.100 --> 01:53:37.542
And when I do it and I eat right, muscles get bigger.

1959
01:53:38.843 --> 01:53:39.503
And here's the thing.

1960
01:53:39.764 --> 01:53:42.726
When you train with an LLM, you're experimenting.

1961
01:53:42.806 --> 01:53:47.429
And this is not actually a hit on the companies, because they're still trying to work out how to do the thing.

1962
01:53:47.650 --> 01:53:47.910
Because...

1963
01:53:48.310 --> 01:53:50.571
Putting aside how I feel like they're trying to innovate.

1964
01:53:50.711 --> 01:53:53.213
I think there are people at these companies that actually want to do something interesting.

1965
01:53:53.513 --> 01:53:54.454
It's costing too much money.

1966
01:53:54.854 --> 01:54:00.778
So once the money tap turns off, the money won't be there to buy the data or feed the data into the GPUs.

1967
01:54:01.118 --> 01:54:07.742
Put aside all the thoughts I have, just the raw capital to get them this far has cost increasingly larger amounts of money.

1968
01:54:08.042 --> 01:54:12.705
And increasingly larger amounts of training money for training runs that sometimes can fail.

1969
01:54:13.265 --> 01:54:17.366
GPT-5 was meant to be this panacea for the AI industry.

1970
01:54:17.606 --> 01:54:20.567
They had at least one training run that cost half a billion dollars and did nothing.

1971
01:54:21.127 --> 01:54:21.847
And that's the thing.

1972
01:54:22.707 --> 01:54:28.608
If we are thinking about progress in a vacuum, they need so much more money just to maybe get somewhere.

1973
01:54:28.628 --> 01:54:29.469
There's no guarantee.

1974
01:54:29.489 --> 01:54:33.109
There's never any guarantee, but there's a reason that Google and Amazon are cash flow negative now.

1975
01:54:33.409 --> 01:54:35.250
There's a reason why Oracle's probably going to die.

1976
01:54:35.810 --> 01:54:41.453
as a result of open AI, because Oracle's future depends on open AI spending $300 billion over five years.

1977
01:54:41.553 --> 01:54:48.837
It's absolutely fascinating, because I was just reading through a list of quotes from the big CEOs of the AI companies to see what they would rebuttal you.

1978
01:54:49.978 --> 01:54:52.199
And they're all basically saying the same thing.

1979
01:54:52.219 --> 01:54:56.921
They're all saying, this is an actual and exact quote from Sundar, who is the CEO of Google.

1980
01:54:57.322 --> 01:55:03.645
He says, the risk of underinvesting is dramatically greater than the risk of overinvesting.

1981
01:55:05.217 --> 01:55:12.720
And you go down, you go through this, you know, Andy Jassy, CEO of Amazon, we're not investing approximately 200 billion in CapEx in 2026 on a hunch.

1982
01:55:13.741 --> 01:55:16.882
We're not going to be conservative in how we play this.

1983
01:55:16.962 --> 01:55:25.186
We're investing to be the meaningful leader, and our future business operating income and free cash flow will be much larger because of this investment.

1984
01:55:25.566 --> 01:55:27.727
Then Mark Zuckerberg, CEO of Meta, says...

1985
01:55:28.267 --> 01:55:31.188
We'll continue to invest aggressively in infrastructure to meet the demand.

1986
01:55:31.548 --> 01:55:35.669
I'd rather risk building capacity before it's needed than being late.

1987
01:55:36.010 --> 01:55:38.130
Makes me think of Shrek with Law Farquad.

1988
01:55:38.430 --> 01:55:41.011
Some of you may die, but that's a risk I'm willing to accept.

1989
01:55:41.051 --> 01:55:43.352
It's like, you know, I'm just going to spend all this money.

1990
01:55:43.512 --> 01:55:48.334
You can't fire me because Mark Zuckerberg can't be fired due to the unique board situation he's got going.

1991
01:55:48.494 --> 01:55:50.574
So yeah, he's just going to piss the money away and hope he's right.

1992
01:55:50.594 --> 01:55:52.015
I don't know from the people who know it matter.

1993
01:55:52.155 --> 01:55:52.795
He's not right.

1994
01:55:53.995 --> 01:55:55.036
Why might you be wrong?

1995
01:55:55.756 --> 01:55:56.577
I mean, this is the thing.

1996
01:55:56.898 --> 01:56:00.882
The AI people who claim this is going to be the biggest, strongest thing in the world, did they ever get that?

1997
01:56:01.063 --> 01:56:01.543
I mean this...

1998
01:56:01.864 --> 01:56:02.584
It's a good question.

1999
01:56:02.725 --> 01:56:03.886
Because it's like, they don't.

2000
01:56:04.126 --> 01:56:06.029
And the thing is, what would it take for me to be wrong?

2001
01:56:06.189 --> 01:56:08.231
A bunch of hardware breakthroughs to make this profitable.

2002
01:56:08.271 --> 01:56:08.572
A bunch of...

2003
01:56:08.592 --> 01:56:09.473
This is a good question, though.

2004
01:56:09.493 --> 01:56:10.694
New mathematics... Because the thing is...

2005
01:56:11.295 --> 01:56:15.339
When it comes to being a critic or a skeptic, you are put on the hot seat.

2006
01:56:15.399 --> 01:56:17.241
Not the people spending a trillion dollars.

2007
01:56:17.281 --> 01:56:19.283
Not the people promising the world.

2008
01:56:19.303 --> 01:56:22.386
The arsehole with a blog is the one who's like, no, no, no.

2009
01:56:22.486 --> 01:56:22.866
Trust me.

2010
01:56:22.906 --> 01:56:24.427
If they came here, they'd be on the hot seat too.

2011
01:56:24.508 --> 01:56:25.408
Trust me.

2012
01:56:26.249 --> 01:56:27.510
Oh, they won't talk to me.

2013
01:56:27.670 --> 01:56:28.371
Don't know why, Steve.

2014
01:56:28.411 --> 01:56:30.773
It's because I call him Clammy Sammy.

2015
01:56:31.154 --> 01:56:31.334
No.

2016
01:56:31.554 --> 01:56:35.677
I think it's because my guests are quite critical that I don't think Sam Altman wants to come here.

2017
01:56:35.777 --> 01:56:37.878
Mr. Altman, go on Steve, do it.

2018
01:56:37.898 --> 01:56:40.120
But this is the thing, like, of course they're going to say that.

2019
01:56:40.220 --> 01:56:43.902
And also, if they thought they were right, I don't think they do anymore.

2020
01:56:44.262 --> 01:56:47.124
If I was in their shoes and I thought that this was an existential thing, sure.

2021
01:56:47.164 --> 01:56:50.847
But it gets back to the rock-con bubble, which is, yeah, this is the last thing they've got.

2022
01:56:51.187 --> 01:56:52.267
But I really want to know that question.

2023
01:56:52.287 --> 01:56:56.048
It was one of the questions I was really excited to ask you, which is you have a different opinion.

2024
01:56:56.069 --> 01:56:56.869
We said this at the top.

2025
01:56:57.049 --> 01:56:59.229
You have a very different opinion from a lot of people.

2026
01:56:59.590 --> 01:57:07.092
I would categorize the two most popular opinions as AI is going to hurt everybody and it's going to be catastrophic and we need to stop.

2027
01:57:07.332 --> 01:57:09.833
The other opinion is age of abundance is going to be amazing.

2028
01:57:09.873 --> 01:57:10.513
Let us crack on.

2029
01:57:10.993 --> 01:57:20.076
Yours is different from both of those, which is, as you said in your words, it's a con and there's no real underlying value in the technology and it's overhyped.

2030
01:57:20.316 --> 01:57:20.536
Yes.

2031
01:57:21.116 --> 01:57:22.457
And there's way too much spending.

2032
01:57:22.477 --> 01:57:24.099
I mean, a few people agree on the spending part.

2033
01:57:24.119 --> 01:57:24.379
Yeah.

2034
01:57:24.559 --> 01:57:25.019
But the other part.

2035
01:57:25.300 --> 01:57:29.683
So with you, it's probably the first person that I've spoken to that's had this opinion.

2036
01:57:30.564 --> 01:57:34.367
So what would it take for you to change your mind?

2037
01:57:35.515 --> 01:57:36.715
About what you believe here.

2038
01:57:37.035 --> 01:57:41.357
There would need to be a hardware breakthrough that reduced the cost by like a thousand.

2039
01:57:41.597 --> 01:57:47.179
It would have to be just a dramatic breakthrough that is not happening, just to be clear, because they've all been trying.

2040
01:57:47.379 --> 01:57:49.139
So it's the cost for you that would have to change.

2041
01:57:49.199 --> 01:57:51.160
It's the cost, and it's also the data centers.

2042
01:57:51.300 --> 01:57:54.481
I think the way they're building the data centers is reckless and damaging to communities.

2043
01:57:54.821 --> 01:58:00.763
The fact that you have communities like in Vinyl, New Jersey, where the residents are like, I don't want this, but the planning boards vote for it because they're all...

2044
01:58:01.303 --> 01:58:04.245
I assume having chummy lunches with the people doing it.

2045
01:58:04.525 --> 01:58:07.208
I think the use of gas turbines is fucking disgraceful.

2046
01:58:07.928 --> 01:58:15.594
The water situation I'm not super well read on, so I'm not going to wade into it, but the use of gas turbines and behind the beat of power is reckless and damaging to communities.

2047
01:58:15.954 --> 01:58:22.039
The noise that these things make, and also, generative AI is this egregious, pornographic...

2048
01:58:23.062 --> 01:58:25.064
demonstration of how unfair the world is.

2049
01:58:25.505 --> 01:58:28.548
Regular people try and get a loan for a business, a random business.

2050
01:58:28.588 --> 01:58:29.569
They want to have a good idea.

2051
01:58:30.169 --> 01:58:32.312
They go to a bank, a bank of talent, they'll go fuck themselves.

2052
01:58:32.352 --> 01:58:35.034
They'll say, you're going to make a store that sells stuff?

2053
01:58:35.114 --> 01:58:35.575
Screw you.

2054
01:58:36.376 --> 01:58:37.637
You want to build a data center?

2055
01:58:37.657 --> 01:58:38.118
The

2056
01:58:38.678 --> 01:58:39.959
Jensen Huang will back you.

2057
01:58:39.979 --> 01:58:42.320
Jensen Huang will give you 25% residual value.

2058
01:58:42.761 --> 01:58:45.583
You want to build a regular business that's even profitable?

2059
01:58:45.663 --> 01:58:46.143
Fuck you.

2060
01:58:46.203 --> 01:58:47.964
No, a venture capitalist won't give you the money.

2061
01:58:48.364 --> 01:58:50.866
Something that's just growing steadily but it's profitable?

2062
01:58:50.906 --> 01:58:51.326
Screw that.

2063
01:58:51.366 --> 01:58:53.448
No, I need 10, 100x return.

2064
01:58:53.888 --> 01:58:54.689
Try and get a mortgage.

2065
01:58:55.069 --> 01:58:57.290
You have to give the bank a full colonic.

2066
01:58:57.731 --> 01:58:59.232
But you want to get money for Jensen Huang?

2067
01:58:59.712 --> 01:59:01.754
To buy some GPUs, he'll give you a contract.

2068
01:59:01.834 --> 01:59:02.875
CoreWeave is a great example.

2069
01:59:03.315 --> 01:59:08.080
A NeoCloud, which is just a company that builds data centers and puts GPUs in rent and people.

2070
01:59:08.400 --> 01:59:12.343
NVIDIA, one of their first investors, in 2023 signed a $1.3 billion contract.

2071
01:59:15.246 --> 01:59:20.148
to rent back their GPUs from CoreWeave so that CoreWeave go to a bank and go, I've got a customer.

2072
01:59:20.568 --> 01:59:24.090
Yeah, it's the guy I'm buying the GPUs from with the debt I'm getting from you.

2073
01:59:24.830 --> 01:59:26.771
If you want to buy GPUs, it's open season.

2074
01:59:26.791 --> 01:59:33.854
If you want to live a regular life where you build a regular business or buy a house, highest interest rates ever, screw you up yours.

2075
01:59:34.174 --> 01:59:36.135
Yeah, you need to show us way more than that.

2076
01:59:36.255 --> 01:59:37.575
I don't trust you regular folks.

2077
01:59:37.876 --> 01:59:42.517
But if you're an unprofitable NeoCloud, you get billions from Jensen.

2078
01:59:42.598 --> 01:59:43.338
It doesn't matter.

2079
01:59:43.838 --> 01:59:44.639
It's so interesting.

2080
01:59:44.659 --> 01:59:48.442
It's interesting because you are the first person that I've spoken to that has that opinion.

2081
01:59:48.802 --> 01:59:49.643
I am pro user.

2082
01:59:50.144 --> 01:59:50.924
Let's take another myth.

2083
01:59:52.286 --> 01:59:53.647
AI will be conscious.

2084
01:59:55.068 --> 01:59:59.051
So, super intelligence, artificial general intelligence.

2085
01:59:59.272 --> 02:00:00.673
These are theories.

2086
02:00:01.534 --> 02:00:06.758
Anyone saying this stuff will become this is just guessing and does not have proof.

2087
02:00:07.059 --> 02:00:07.759
Okay.

2088
02:00:08.079 --> 02:00:08.740
And that's really it.

2089
02:00:08.980 --> 02:00:09.160
Okay.

2090
02:00:09.881 --> 02:00:10.862
Okay, let's take another myth.

2091
02:00:12.644 --> 02:00:16.225
AI systems are already blackmailing and escaping control.

2092
02:00:16.426 --> 02:00:17.786
So this is a really specific one.

2093
02:00:18.046 --> 02:00:18.646
Anthropic.

2094
02:00:19.107 --> 02:00:19.707
There's actually two.

2095
02:00:20.527 --> 02:00:22.308
OpenAI's GPT 3.5.

2096
02:00:23.048 --> 02:00:24.309
I realize this is more than a sentence.

2097
02:00:24.349 --> 02:00:24.949
I apologize.

2098
02:00:26.149 --> 02:00:33.832
In their system card, and a bunch of media outlets cover this, saying that OpenAI's model blackmailed a TaskRabbit into solving a capture.

2099
02:00:34.252 --> 02:00:40.375
What actually happened was a user of GPT doing the experiment

2100
02:00:41.798 --> 02:00:46.401
got it to generate things to say to a TaskRabbit to make a TaskRabbit do stuff.

2101
02:00:46.441 --> 02:00:50.103
A TaskRabbit being... As in a person that you rent, not even to do a capture.

2102
02:00:50.143 --> 02:00:53.264
It's someone you rent to nail a picture up in your apartment.

2103
02:00:53.284 --> 02:00:54.365
It's an insane example.

2104
02:00:54.745 --> 02:00:57.507
This was covered as if these things blackmailed someone.

2105
02:00:57.527 --> 02:01:00.829
And they specifically said, yeah, we prompted it to do this.

2106
02:01:01.169 --> 02:01:05.471
And also, the other note was that, yeah, AI systems can't do autonomous stuff like this.

2107
02:01:05.991 --> 02:01:08.413
Then there was this other one where Anthropic said, oh, yeah...

2108
02:01:08.993 --> 02:01:16.181
A model was blackmailing someone, saying that if you don't do this, I'll email proof that you slept with someone else other than your wife, I think it was.

2109
02:01:16.582 --> 02:01:22.128
What actually happened was Anthropic explicitly trained a model to do this and then prompted it to blackmail.

2110
02:01:23.358 --> 02:01:27.160
This keeps happening and the media just slop me up.

2111
02:01:27.260 --> 02:01:28.461
I don't need no thoughts.

2112
02:01:28.841 --> 02:01:30.062
Put the story in the bag.

2113
02:01:30.362 --> 02:01:32.903
And it's frustrating because it scares people.

2114
02:01:33.083 --> 02:01:34.504
Put aside the fact it's wrong.

2115
02:01:34.764 --> 02:01:35.605
It's scary.

2116
02:01:35.805 --> 02:01:42.208
It's scary to people, people living their lives who have to work longer hours to make less money and their money doesn't go far.

2117
02:01:42.368 --> 02:01:46.230
And they turn on the fucking news and there's some asshole being like, yeah, you should be terrified.

2118
02:01:46.250 --> 02:01:47.291
It blackmailed someone.

2119
02:01:47.931 --> 02:01:52.037
But this is so counterintuitive of their interest to some degree.

2120
02:01:52.237 --> 02:01:54.020
And they've experienced it backfire.

2121
02:01:54.280 --> 02:01:54.921
Well, they have now.

2122
02:01:55.162 --> 02:01:56.323
Like, it's literally backfired.

2123
02:01:56.343 --> 02:01:57.145
It's backfired.

2124
02:01:57.225 --> 02:02:01.731
Eric Schmidt getting booed at the commencement speech by 8,000 people every time he said the word AI.

2125
02:02:01.751 --> 02:02:02.212
Yeah, that's awesome.

2126
02:02:02.873 --> 02:02:06.634
But I mean, these CEOs are being attacked at home.

2127
02:02:07.054 --> 02:02:08.275
Yeah, which fucking sucks.

2128
02:02:08.295 --> 02:02:09.215
Yeah, which is terrible.

2129
02:02:09.335 --> 02:02:12.296
I must be clear, like, you dislike the company, don't fucking hurt people.

2130
02:02:12.356 --> 02:02:13.576
Yeah, don't attack people at home.

2131
02:02:13.816 --> 02:02:19.158
But the point here is that that narrative is backfiring in a big, big way for them.

2132
02:02:19.298 --> 02:02:20.739
I don't think they saw it coming.

2133
02:02:20.859 --> 02:02:23.240
Because you have to remember, you mentioned regulation earlier.

2134
02:02:23.581 --> 02:02:27.103
These tech companies have been glazed for their entire existence.

2135
02:02:27.343 --> 02:02:28.904
Travis Kalanick's like, oh, what?

2136
02:02:28.964 --> 02:02:29.884
People don't like me now.

2137
02:02:29.904 --> 02:02:33.406
And it's because Uber was a horribly run place and he was kind of a monster.

2138
02:02:33.927 --> 02:02:36.228
Also, tons of articles about how great Uber was at the time.

2139
02:02:36.308 --> 02:02:38.509
The point I'm making is these companies are not used to pushback.

2140
02:02:39.050 --> 02:02:48.155
They thought what would happen, I believe, just guessing, they thought they'd do this scary stuff and they would just get floods of money and everyone would just be like, I kneel before you, I'll do whatever you want.

2141
02:02:49.195 --> 02:02:51.656
They didn't expect, I think, what has happened.

2142
02:02:52.016 --> 02:02:53.696
I agree, this has backfired on them.

2143
02:02:54.076 --> 02:02:55.637
Because they were inarticular.

2144
02:02:55.837 --> 02:02:57.977
They're disconnected from regular people.

2145
02:02:58.017 --> 02:03:01.398
Sam Altman drives a $5 million car around San Francisco.

2146
02:03:01.418 --> 02:03:03.618
So that man's doing it like nine miles an hour.

2147
02:03:03.658 --> 02:03:04.219
It's hilarious.

2148
02:03:04.679 --> 02:03:06.719
But these people are disconnected from everyone else.

2149
02:03:06.799 --> 02:03:10.280
So they don't experience real problems, so they can't build the solutions for them.

2150
02:03:10.820 --> 02:03:13.981
And they think, well, if we scare people into doing what we want, that'll work, right?

2151
02:03:14.541 --> 02:03:14.881
It didn't.

2152
02:03:15.461 --> 02:03:18.982
This was, all of this blackmail stuff, was an attempt to make it mystic.

2153
02:03:19.422 --> 02:03:20.483
It was a mysticism attempt.

2154
02:03:20.523 --> 02:03:24.906
It was to make it seem like this unknowable, impossible to control, just this powerful thing.

2155
02:03:25.006 --> 02:03:26.006
But we're the only ones.

2156
02:03:26.647 --> 02:03:28.228
We are the only us.

2157
02:03:28.708 --> 02:03:32.771
Only these two angels could possibly control the beast we've created.

2158
02:03:33.331 --> 02:03:34.972
This is quite a controversial statement.

2159
02:03:35.012 --> 02:03:38.975
But I think that for some reason, I trust Dario a little bit more.

2160
02:03:39.873 --> 02:03:44.214
Because I think he's been the most balanced in his writing about the risk profile.

2161
02:03:45.314 --> 02:03:51.096
Whereas the others, they seem to kind of move with the wind.

2162
02:03:51.196 --> 02:03:51.976
I get what you mean.

2163
02:03:52.356 --> 02:03:56.977
The reason I don't like Dario is Dario was doing the scare tactics thing when he worked at OpenAI.

2164
02:03:57.157 --> 02:03:59.818
When GPT-2 came out, he said, it's too scary to release.

2165
02:04:00.098 --> 02:04:07.580
He's also gone on television and given AI psychosis to Axios, being like, 50% of jobs are going to go away because of AI.

2166
02:04:07.980 --> 02:04:11.102
What I respect is the consistency.

2167
02:04:11.362 --> 02:04:12.722
He's now being attacked by them.

2168
02:04:13.203 --> 02:04:13.483
Good.

2169
02:04:14.263 --> 02:04:15.064
But the thing is...

2170
02:04:15.084 --> 02:04:17.145
Sorry, I mean, clarify the word attack.

2171
02:04:17.525 --> 02:04:20.506
Dario is being verbally attacked by Silicon Valley.

2172
02:04:20.866 --> 02:04:25.889
And, you know, if Silicon Valley, if powerful people in Silicon Valley are attacking someone...

2173
02:04:26.389 --> 02:04:27.810
Four months ago, he wasn't, though.

2174
02:04:27.830 --> 02:04:29.611
They were all saying he was the smartest boy ever.

2175
02:04:29.692 --> 02:04:33.254
But the point I want to make there as well is, again, wow, you're so scared of how powerful this is.

2176
02:04:33.274 --> 02:04:34.255
You're so scared of it.

2177
02:04:34.275 --> 02:04:34.915
It's so scary.

2178
02:04:34.935 --> 02:04:35.756
What are you doing about it?

2179
02:04:35.856 --> 02:04:37.157
Oh, nothing.

2180
02:04:37.177 --> 02:04:38.298
It's just like, what are you doing?

2181
02:04:38.678 --> 02:04:39.719
Well, we have an alignment team.

2182
02:04:39.919 --> 02:04:41.020
So does every AI lab.

2183
02:04:41.180 --> 02:04:43.702
Well, I guess OpenAI cycles through this really quickly.

2184
02:04:44.102 --> 02:04:44.542
Here's the thing.

2185
02:04:44.582 --> 02:04:47.864
If I'm Dario Amadeus and I'm like, I'm scared of all things changing...

2186
02:04:48.685 --> 02:04:51.428
And I thought I had made a thing that would eliminate all jobs.

2187
02:04:51.788 --> 02:04:53.029
I'd be fucking terrified.

2188
02:04:53.270 --> 02:04:57.534
I'd be walking around with like a 10-ton weight on my back.

2189
02:04:57.594 --> 02:05:08.004
The responsibility, the fact he doesn't, the fact he wants to be this weird elder statesman that's too scared to hold Sam Altman's hand at an event just makes me believe that he's just saying it because it's convenient.

2190
02:05:08.224 --> 02:05:10.427
And he'll wind that back as he kind of already has...

2191
02:05:10.907 --> 02:05:12.147
whenever it's convenient for him.

2192
02:05:12.448 --> 02:05:15.569
I think OpenAI and Anthropic are basically the same level of bad company.

2193
02:05:15.829 --> 02:05:18.090
I think Anthropic is more cult-like.

2194
02:05:18.470 --> 02:05:30.334
I think it's so weird, like Jack Clark over there, one of the co-founders, that fellow used to be at the register, he used to be one of the most critical journalists ever, and now it's like something took over him, because they talk of these things in these highfalutin terms.

2195
02:05:30.454 --> 02:05:35.056
But then again, maybe the people at Anthropic buy their shit, maybe some of the people at OpenAI buy their shit, I don't know.

2196
02:05:35.076 --> 02:05:37.137
So going back to the central question we asked at the top here was,

2197
02:05:37.957 --> 02:05:41.600
What would have to be the case for you to look back and say, do you know what, I was wrong in 2026.

2198
02:05:41.840 --> 02:05:48.746
And you said to me it would be mainly that the cost of production around AI drops dramatically.

2199
02:05:49.026 --> 02:05:52.288
And it would have to also do insane amounts of stuff it does.

2200
02:05:52.328 --> 02:05:54.470
It would have to be a truly autonomous...

2201
02:05:54.490 --> 02:05:57.312
It would have to continue its improvement in terms of capability.

2202
02:05:57.332 --> 02:05:59.114
It would have to be a different product.

2203
02:05:59.214 --> 02:06:01.095
It would have to be indistinguishable from magic.

2204
02:06:01.115 --> 02:06:03.197
And the reason I have these high standards is they set them.

2205
02:06:03.557 --> 02:06:04.138
Okay, fair enough.

2206
02:06:04.158 --> 02:06:10.503
It's interesting as well because all these myths and all these conversations, it's about technology, but it's also, it's an information war.

2207
02:06:10.943 --> 02:06:11.604
It's literally...

2208
02:06:12.844 --> 02:06:31.860
narrative versus narrative everyone trying to escape the financials everyone trying to actually escape what the models can do and the big thing i always say about ai boosters is if i could regulate them i'd regulate them they can't speak in the future tense anymore just you got to talk about today mate you get two weeks in the future max because if they were constrained to what was

2209
02:06:35.403 --> 02:06:36.524
Yeah, no, I think...

2210
02:06:36.664 --> 02:06:38.725
I guess most technology companies would at the time.

2211
02:06:38.845 --> 02:06:39.965
Uber would sound insane.

2212
02:06:40.105 --> 02:06:41.066
Amazon would sound insane.

2213
02:06:41.166 --> 02:06:42.566
Uber was basically...

2214
02:06:42.627 --> 02:06:43.787
They were pissing money, though, weren't they?

2215
02:06:43.807 --> 02:06:47.289
They were pissing money away, but the unit economics were the same, just subsidized.

2216
02:06:47.309 --> 02:06:52.571
So you were still getting a service from A to B and paying a much lower cost.

2217
02:06:52.911 --> 02:06:53.672
It wasn't like...

2218
02:06:54.212 --> 02:06:58.495
You paid Uber 200, sorry, 20 bucks a month and you could get 500 miles of Uber.

2219
02:06:58.715 --> 02:07:00.376
And then one day you started paying by the mile.

2220
02:07:00.396 --> 02:07:01.497
Because that's what's happening with this.

2221
02:07:01.697 --> 02:07:07.301
Have they, they've changed their business model for customers like me now so that I have to buy credits?

2222
02:07:07.521 --> 02:07:07.762
No.

2223
02:07:07.882 --> 02:07:09.583
So you, well, kind of with Fable.

2224
02:07:09.603 --> 02:07:10.904
They asked me the other day to buy credit.

2225
02:07:10.924 --> 02:07:15.767
So with the Anthropics Fable model, with some accounts you have to pay for usage.

2226
02:07:16.007 --> 02:07:19.250
And also adoption of Fable has been pretty low because of this, because of the cost.

2227
02:07:19.770 --> 02:07:25.931
But with enterprises, so companies over 150 people, you have to pay by the token now, or per million token.

2228
02:07:26.352 --> 02:07:27.372
Oh, so they are moving to a token.

2229
02:07:27.392 --> 02:07:34.854
Yeah, but when they did that, everyone went from being like, this is the most impressive thing ever to being like, we've got to control these costs.

2230
02:07:35.034 --> 02:07:44.016
Uber's COO says, Andrew McDonald, I think he said that it's getting hard to justify because it's hard to connect spending money on tokens to actual useful outcomes.

2231
02:07:44.676 --> 02:07:45.516
He said the thing.

2232
02:07:45.716 --> 02:07:48.457
Like, he said the actual thing I've been saying, and it's...

2233
02:07:48.477 --> 02:07:49.497
So we're in an AI bubble.

2234
02:07:49.717 --> 02:07:49.917
Yes.

2235
02:07:50.378 --> 02:07:55.021
And when this AI bubble collapses, so much of the economy is resting upon it.

2236
02:07:56.102 --> 02:07:56.302
Yeah.

2237
02:07:56.522 --> 02:07:58.164
It's going to have downstream consequences.

2238
02:07:58.184 --> 02:07:59.264
So I've got two questions for you.

2239
02:07:59.605 --> 02:08:01.406
I guess the first question is, are we in an AI bubble?

2240
02:08:01.806 --> 02:08:03.187
And what happens when the bubble pops?

2241
02:08:03.688 --> 02:08:06.350
Yes, and it depends.

2242
02:08:06.610 --> 02:08:10.093
So the big thing that people say is, oh, we'll get bailed out.

2243
02:08:10.133 --> 02:08:11.634
Donald Trump, scared of Donald Trump.

2244
02:08:11.974 --> 02:08:12.895
Here's the problem with this.

2245
02:08:14.520 --> 02:08:17.382
It isn't just an AI bubble, it's the Rotcom bubble.

2246
02:08:17.542 --> 02:08:22.065
So the AI bubble collapsing will probably be this company running out of money.

2247
02:08:22.526 --> 02:08:23.166
Open AI.

2248
02:08:23.847 --> 02:08:26.909
And the thing is with Open AI is they were meant to go public this year.

2249
02:08:27.309 --> 02:08:28.630
And now it's been pushed to next year.

2250
02:08:28.950 --> 02:08:31.072
A week and a half after I released their auditive financials.

2251
02:08:31.132 --> 02:08:31.812
I wonder where that was.

2252
02:08:32.453 --> 02:08:33.754
But they've delayed to next year.

2253
02:08:33.794 --> 02:08:35.335
Sarah Fryer, the CFO, has now said...

2254
02:08:35.895 --> 02:08:39.536
Well, they'll do it earlier than 2027 or 2027.

2255
02:08:39.856 --> 02:08:40.616
Great answer there.

2256
02:08:40.876 --> 02:08:44.397
For anyone that doesn't understand what going public means, that means joining the stock market.

2257
02:08:44.677 --> 02:08:52.738
And at such a time when you join the stock market, your investors can finally sell their equity that they got for investing in the company when it was private.

2258
02:08:53.218 --> 02:09:03.260
So oftentimes companies will flirt with the idea of we'll go public someday soon because investors will have a moment in their head where they'll get their money back at a return.

2259
02:09:03.960 --> 02:09:09.284
So you kind of need to, if you're in these guys' shoes, you kind of need to be flirting with going public or investors won't want to invest.

2260
02:09:10.005 --> 02:09:12.547
OpenAI up until this point has been a private company.

2261
02:09:12.827 --> 02:09:16.530
And their last funding round, they were valued at $865 billion.

2262
02:09:17.270 --> 02:09:21.293
Now, when they tried to go public, New York Times' Mike Isaac reported this.

2263
02:09:22.094 --> 02:09:26.740
They tried to list, well, they wanted to go at a $1 trillion valuation.

2264
02:09:27.441 --> 02:09:30.765
Apparently, their advisor said, no, don't do that.

2265
02:09:31.566 --> 02:09:33.529
That is very bad for a number of reasons.

2266
02:09:33.649 --> 02:09:35.572
One, OpenAI needs perpetual amounts of money.

2267
02:09:35.592 --> 02:09:37.534
They raised $122 billion this year.

2268
02:09:37.775 --> 02:09:38.656
Most of it's crossed.

2269
02:09:38.676 --> 02:09:39.417
There's some left.

2270
02:09:39.577 --> 02:09:39.817
But...

2271
02:09:40.538 --> 02:09:43.600
They are going to need to raise at least $100 billion a year just to survive.

2272
02:09:44.260 --> 02:09:46.981
If they can't go public, they will have to raise another funding round.

2273
02:09:47.302 --> 02:09:51.484
The problem is it's going to be difficult to raise even the same one they raised that.

2274
02:09:51.504 --> 02:09:53.865
They're probably going to have to take a flat, so the same amount.

2275
02:09:53.985 --> 02:09:54.705
They can't do that.

2276
02:09:55.026 --> 02:09:55.546
Exactly.

2277
02:09:56.126 --> 02:09:56.767
But they need money.

2278
02:09:56.967 --> 02:09:58.087
They need money so bad.

2279
02:09:58.207 --> 02:10:03.250
Amazon sent them $35 billion that was meant to be contingent on them going public early.

2280
02:10:04.195 --> 02:10:05.596
They did that because they need the money.

2281
02:10:06.056 --> 02:10:12.199
Now, OpenAI is the kind of catastrophe center here because Anthropic is likely going to beat it to go public.

2282
02:10:12.279 --> 02:10:22.043
And once Anthropic goes public, it'll be borderline impossible for OpenAI to do so because Anthropic, an unprofitable, unsustainable AI lab, but a better business that's growing faster than OpenAI's.

2283
02:10:22.463 --> 02:10:23.424
I believe they have a ceiling.

2284
02:10:23.584 --> 02:10:25.125
They're eventually going to face perdition too.

2285
02:10:25.865 --> 02:10:28.848
I think sometime in 2027, things are going to start running out of steam.

2286
02:10:28.868 --> 02:10:34.454
Because the thing I said earlier, the only way these models get better is if you feed more money, tens of billions of dollars into them.

2287
02:10:34.915 --> 02:10:37.278
So you think OpenAI runs out of steam in 2027?

2288
02:10:37.718 --> 02:10:39.340
I think they're already running out of steam, yeah.

2289
02:10:39.360 --> 02:10:40.541
But I think they run out of cash.

2290
02:10:40.781 --> 02:10:42.043
You think they run out of cash?

2291
02:10:42.203 --> 02:10:42.443
Yes.

2292
02:10:42.583 --> 02:10:45.567
And the sequence of events here will be they go out and try and raise...

2293
02:10:46.107 --> 02:10:47.828
And they have trouble raising another round.

2294
02:10:47.928 --> 02:10:50.130
I think maybe NVIDIA props them up a little.

2295
02:10:50.150 --> 02:10:53.492
Maybe Private Credit, Blackstone, BlackRock, and the like.

2296
02:10:53.652 --> 02:11:00.857
And the reason that Private Credit is getting involved, so asset managers, is because they're investing in the data centers, and they know this company's most of the data center demand.

2297
02:11:01.097 --> 02:11:03.419
Okay, so they run out of steam in 2027, according to you.

2298
02:11:03.599 --> 02:11:03.759
Yep.

2299
02:11:03.899 --> 02:11:07.742
And maybe they try, if they bum rush to go public, they're going to have worse economics than an anthropic.

2300
02:11:07.782 --> 02:11:08.963
They're going to get savage.

2301
02:11:09.343 --> 02:11:10.624
WeWork was a great example.

2302
02:11:10.984 --> 02:11:12.145
Another SoftBank classic.

2303
02:11:12.957 --> 02:11:14.678
Now, I think OpenAI collapses.

2304
02:11:14.898 --> 02:11:16.479
There are many different ways it could happen.

2305
02:11:16.759 --> 02:11:18.120
There are many different ways it could end.

2306
02:11:18.840 --> 02:11:24.144
But the crucial thing is that there are multiple companies that are existentially tied to OpenAI.

2307
02:11:24.624 --> 02:11:31.388
SoftBank, one of the largest companies in the Japanese stock market, a holding company with lots of investments, they have on paper...

2308
02:11:31.948 --> 02:11:34.209
about $100 billion worth of OpenAI stock.

2309
02:11:34.710 --> 02:11:37.211
If they can't go public, they can't do diddly squat with them.

2310
02:11:37.772 --> 02:11:52.761
And so SoftBank's future, their ability to continue paying the people around them and existing as a business, relies on their ability to continually liquidate funds, to take the things they've invested in and have value from them, either by selling the stock or taking loans out on the stock.

2311
02:11:53.141 --> 02:11:55.942
If OpenAI can't go public, SoftBank can't do that.

2312
02:11:56.022 --> 02:11:56.463
SoftBank.

2313
02:11:57.003 --> 02:12:01.748
Probably won't run out of money, but we're going to see one of the largest holding companies in the world become much smaller.

2314
02:12:02.288 --> 02:12:07.133
We will also see Amazon, Google and Microsoft have to restate guidance.

2315
02:12:07.253 --> 02:12:10.055
They will have to say, actually, we don't think we're going to grow as fast.

2316
02:12:10.656 --> 02:12:11.337
And what happens then?

2317
02:12:11.996 --> 02:12:20.523
Well, I think we enter a tech depression because the rock-com bubble, the core of my theory, is that they're out of hyper-growth ideas, but the market doesn't think so.

2318
02:12:20.543 --> 02:12:27.969
The reason they're so maniacally spending is because buying AI GPUs allows them to kick the can further.

2319
02:12:27.989 --> 02:12:29.911
It allows them to say, we're still doing something.

2320
02:12:29.971 --> 02:12:30.771
We're working on AI.

2321
02:12:31.031 --> 02:12:31.872
Don't think too hard.

2322
02:12:32.092 --> 02:12:33.714
And also, their current business is still growing.

2323
02:12:33.734 --> 02:12:34.094
Yeah.

2324
02:12:34.369 --> 02:12:36.250
their current businesses will eventually slow.

2325
02:12:36.450 --> 02:12:38.050
There's only so many price increases.

2326
02:12:38.070 --> 02:12:39.651
There's only so many tweaks to ads.

2327
02:12:40.031 --> 02:12:42.112
Only so many tweaks to Google search.

2328
02:12:42.592 --> 02:12:45.053
Only so many ways that Amazon can screw merchants.

2329
02:12:45.473 --> 02:12:54.737
So in that tech depression, which you think it might be triggered in 2027, is that a cascading downstream economic depression?

2330
02:12:54.877 --> 02:12:58.018
Because the stock market is heavily dependent on these companies.

2331
02:12:58.058 --> 02:12:59.299
The stock market sees a pullback.

2332
02:12:59.719 --> 02:13:00.900
Investors stop investing.

2333
02:13:00.920 --> 02:13:01.720
They get panicked.

2334
02:13:02.573 --> 02:13:02.873
Yes.

2335
02:13:03.253 --> 02:13:06.614
I think that because... What sort of downstream consequence, the sort of domino effect?

2336
02:13:06.634 --> 02:13:12.116
There's so much to imagine that it's difficult to capture everything, but there are a few things that worry me.

2337
02:13:12.156 --> 02:13:17.337
First of all, a ton of American money, just regular people's money, retail investors, are in these companies.

2338
02:13:17.357 --> 02:13:20.178
And they bought into the Magnificent Seven thinking that number go up forever.

2339
02:13:20.738 --> 02:13:24.219
NVIDIA is the largest company on the Fortune 500 and NASDAQ as well.

2340
02:13:24.859 --> 02:13:26.139
And like 7% to 8% of the S&P 500...

2341
02:13:28.402 --> 02:13:37.387
That company, when the bottom falls out from NVIDIA, and we haven't really got into it, but NVIDIA is doing the most circular of financing, feeding companies money so that they can raise debt to buy more GPUs.

2342
02:13:38.248 --> 02:13:41.289
I think NVIDIA's revenue could go 50% to 70% now.

2343
02:13:41.509 --> 02:13:45.492
I think that NVIDIA could, NVIDIA back in 2022 was making single digit billion dollars.

2344
02:13:45.852 --> 02:13:51.095
And what happens though, I'm thinking about like Jenny and Dave that are watching this right now, and they are just normal people.

2345
02:13:51.859 --> 02:13:52.580
With normal jobs.

2346
02:13:52.600 --> 02:13:58.185
People's retirements are going to contract severely, and I don't believe they're going to return to those values.

2347
02:13:58.485 --> 02:14:08.834
And I think that because so much of the value of the S&P 500 and Russell 1000 index comes from these four companies and the rest of the Magnificent Seven, so Apple, Tesla, Meta as well.

2348
02:14:09.716 --> 02:14:17.526
And the thing is, I don't know what happens after that because venture capital has also, like more than half of venture capital last year went into AI.

2349
02:14:17.907 --> 02:14:20.650
I think most venture capital investments in AI are going to zero.

2350
02:14:20.910 --> 02:14:25.116
Because when it comes to building a company on top of an LLM, all of those are unprofitable too.

2351
02:14:25.677 --> 02:14:26.217
And the thing is,

2352
02:14:27.218 --> 02:14:32.361
LLM companies have not really been acquired, the exception being Cursable by Elon Musk for the coding side.

2353
02:14:32.781 --> 02:14:38.604
But you have Cognition, which is just another LLM company raising a $26 billion valuation.

2354
02:14:38.964 --> 02:14:43.566
That means that company has to go public because who's buying a company at $26 billion other than Elon Musk?

2355
02:14:43.967 --> 02:14:46.148
And there are rumors that Elon Musk was trying to buy them as well.

2356
02:14:46.448 --> 02:14:48.329
Is Elon Musk just going to pick off every LLM?

2357
02:14:49.049 --> 02:14:51.971
Like LLM company, like on a fucking TJ Maxx for AI, like Jesus Christ.

2358
02:14:52.271 --> 02:14:53.832
So is that a recession you're describing?

2359
02:14:54.212 --> 02:14:57.894
It is a recession, but it's also a depression within people's retirements.

2360
02:14:57.914 --> 02:15:02.517
Like I'm talking about 20, 30, 40% off the top of these companies' stock value.

2361
02:15:02.597 --> 02:15:06.599
Economic contractions, recessions consistently lead to job losses and rising unemployment.

2362
02:15:06.659 --> 02:15:10.541
When an economy contracts, the mechanism driving job losses typically follows a predictable sequence.

2363
02:15:11.642 --> 02:15:16.005
Falling demand, consumers and businesses spend less money causing revenues across most industries to drop.

2364
02:15:16.145 --> 02:15:22.248
Margin compression with lower revenue and often fixed overhead costs like rent or debt, corporate profit shrink.

2365
02:15:22.308 --> 02:15:28.212
And lastly, cost cutting measures to survive or protect profit margins, businesses freeze hiring, reduce hours and resort to layoffs.

2366
02:15:28.877 --> 02:15:30.598
Yes, that would all happen.

2367
02:15:30.638 --> 02:15:38.322
But the thing is, we're talking about equity values dropping, and we're talking about there not really being a home for that value or that money.

2368
02:15:39.182 --> 02:15:42.544
So much is riding on these companies, but you can't bail it out.

2369
02:15:43.024 --> 02:15:45.145
You can theoretically bail out OpenAI.

2370
02:15:45.165 --> 02:15:45.906
I don't think it happens.

2371
02:15:45.946 --> 02:15:51.028
You could pump these dogs full of money and keep them alive for a bit, but at some point they're going to have to start.

2372
02:15:51.268 --> 02:15:53.810
They have, between these two companies, Anthropic and OpenAI, a lot of companies.

2373
02:15:54.682 --> 02:15:57.104
you have $1.1 trillion of commitments.

2374
02:15:57.664 --> 02:15:58.525
Just OpenAI.

2375
02:15:59.246 --> 02:16:05.190
Oracle is building 7.1 gigawatts of data centers, so over $400 billion worth just for OpenAI.

2376
02:16:05.390 --> 02:16:10.494
There is not a customer on earth, and Oracle's revenue has been flat the last 15 years when you adjust for inflation.

2377
02:16:10.655 --> 02:16:12.656
Without OpenAI, Oracle dies.

2378
02:16:12.776 --> 02:16:21.483
So you think OpenAI is going to crash and run out of money, and that's going to cause this domino effect across these other big tech companies, which is going to impact the stock market and impact the broader economy?

2379
02:16:21.563 --> 02:16:25.285
Yes, and also the tens of thousands of people that will be laid off from the tech sector.

2380
02:16:25.325 --> 02:16:34.809
But also, the venture capital thing is significant because venture capital has been having one of the most historic bad runs in history.

2381
02:16:35.209 --> 02:16:46.553
Since 2018, the average return from venture capital, a total value put in, so the amount of money you get back for your dollar is between 0.8 and 1.21, meaning for every dollar you invest, you get 80 cents to $1.20.

2382
02:16:46.874 --> 02:16:47.394
Paper gains.

2383
02:16:48.094 --> 02:16:50.376
Well, no, that's just actual returns.

2384
02:16:50.416 --> 02:16:55.562
Paper gains they'll give you, but even an internal rate return, which is a whole separate thing, even that's not very happy.

2385
02:16:55.622 --> 02:16:59.966
But long story short, very simple, venture capital is not making money come out.

2386
02:17:00.387 --> 02:17:03.089
Venture capital is not actually providing returns.

2387
02:17:03.289 --> 02:17:04.531
They're celebrating paper gains.

2388
02:17:04.871 --> 02:17:05.812
They're celebrating paper gains.

2389
02:17:05.852 --> 02:17:06.933
And they're raising off paper gains.

2390
02:17:07.013 --> 02:17:07.373
Uh-huh.

2391
02:17:07.693 --> 02:17:11.977
And paper gains, I mean, just being able to say, oh, look, the valuation of Anthropic went up.

2392
02:17:12.017 --> 02:17:13.978
But that's what Google and Amazon were doing.

2393
02:17:14.018 --> 02:17:24.066
Google's last quarter, they boosted their net profits, profits on paper, by $99 billion because of the increased value of their SpaceX holding and their Anthropic holding.

2394
02:17:24.526 --> 02:17:25.187
And again...

2395
02:17:26.343 --> 02:17:31.947
The fact that this is happening is insane, and the fact it's not a scandal is insane, but we live in this culture, I guess.

2396
02:17:32.627 --> 02:17:35.829
But everyone is really benefiting right now.

2397
02:17:35.910 --> 02:17:41.794
It's really that great tweet where it's like, when you're reaping, it's like, yeah, fuck yeah, this rocks, sewing, ah, shit, this sucks.

2398
02:17:41.854 --> 02:17:51.680
Because right now they're all like, yeah, all the speculative gains are awesome, the paper gains are awesome, the theoreticals of Anthropic being worth $2 trillion, wow, the articles we can write, the promises we can make.

2399
02:17:52.261 --> 02:17:53.742
Then when the rubber meets the road...

2400
02:17:54.488 --> 02:18:03.055
It's going to be pretty rough on them because the valuation of Amazon, Google, Microsoft and Meta is based on this idea that they will grow eternally, that they will grow forever.

2401
02:18:03.956 --> 02:18:08.819
If that changes, to quote Ed Elson from Prof G Markets again, it's this, they're all doing Botox right now.

2402
02:18:08.859 --> 02:18:11.101
They're sinking money into it to make themselves feel young again.

2403
02:18:11.281 --> 02:18:12.322
And the market believes them.

2404
02:18:12.602 --> 02:18:15.004
When the market doesn't, we're not just talking about a depression.

2405
02:18:15.365 --> 02:18:21.690
I'm talking about the market valuing them like airlines and saying, yeah, you're real big and you make money off your existing products.

2406
02:18:22.290 --> 02:18:22.951
But guess what?

2407
02:18:23.211 --> 02:18:24.171
You don't have new shit.

2408
02:18:24.452 --> 02:18:25.913
You're just going to be doing this forever.

2409
02:18:26.093 --> 02:18:27.634
And we're going to value you as such.

2410
02:18:27.794 --> 02:18:31.516
So for Jenny and Dave, should they do anything differently?

2411
02:18:31.997 --> 02:18:33.117
Should they be conserving money?

2412
02:18:33.157 --> 02:18:35.999
If there's a recession or depression coming, should they be a little bit more conservative?

2413
02:18:36.019 --> 02:18:36.580
Should they?

2414
02:18:36.760 --> 02:18:37.060
Yes.

2415
02:18:37.340 --> 02:18:39.882
I actually think it's, I don't know.

2416
02:18:40.022 --> 02:18:41.063
I don't have money in the market.

2417
02:18:41.103 --> 02:18:42.084
I think it's a casino.

2418
02:18:42.244 --> 02:18:43.905
Casino pumped up by the media.

2419
02:18:44.225 --> 02:18:45.406
Should they invest in the S&P 500?

2420
02:18:45.426 --> 02:18:46.846
Should they invest in OpenAI?

2421
02:18:47.287 --> 02:18:48.067
Oh, God, no.

2422
02:18:48.487 --> 02:18:50.048
Honestly, I live in cash right now.

2423
02:18:50.348 --> 02:18:50.949
You live in cash?

2424
02:18:50.989 --> 02:18:51.169
Yeah.

2425
02:18:51.369 --> 02:18:52.770
I don't fucking trust the market, man.

2426
02:18:53.050 --> 02:18:55.151
Try and get some gains here.

2427
02:18:55.211 --> 02:19:00.254
I'm not comfortable giving financial advice, but it's like you're gambling.

2428
02:19:00.414 --> 02:19:01.374
Okay, be conservative.

2429
02:19:01.414 --> 02:19:02.295
Things might get volatile.

2430
02:19:02.675 --> 02:19:03.435
Yeah, it really is.

2431
02:19:03.535 --> 02:19:05.396
It's going to be act as you would with volatility.

2432
02:19:05.476 --> 02:19:06.777
Take the gains when you've got them.

2433
02:19:07.397 --> 02:19:10.878
Don't sell everything, but be suspicious of tech.

2434
02:19:10.938 --> 02:19:11.959
Like, that's actually the biggest thing.

2435
02:19:11.979 --> 02:19:13.779
It's like, be suspicious of what they're promising.

2436
02:19:13.819 --> 02:19:17.861
If you're acting based on their promises, don't trust the promises.

2437
02:19:18.521 --> 02:19:31.745
Trust that they are going to say what will make the stock run rather than what's actually happening and that they will find every dodgy way to make you think something is happening rather than it's actually happening.

2438
02:19:32.045 --> 02:19:32.865
Annualized run rate.

2439
02:19:32.905 --> 02:19:33.526
Great example.

2440
02:19:34.186 --> 02:19:38.667
Microsoft said that they had $38, $37 billion of annualized run rate in AI.

2441
02:19:38.867 --> 02:19:42.729
You hear that, you go, I made $38, $37 billion, right?

2442
02:19:42.849 --> 02:19:43.509
Wow, that's so much.

2443
02:19:44.309 --> 02:19:46.451
Run rate may be month times 12.

2444
02:19:46.551 --> 02:19:49.193
They don't even define it, but it's built to manipulate.

2445
02:19:49.393 --> 02:19:58.980
And they do that because we don't have a functional SEC and we don't have a media environment that actually where skepticism is the priority and where protecting the readers is necessary.

2446
02:19:59.081 --> 02:19:59.721
What would they say?

2447
02:19:59.741 --> 02:20:12.271
They would say, Ed, this technology is going to be so great and so transformative that we are investing a ton of money in advance of the value and utility showing up.

2448
02:20:12.371 --> 02:20:13.452
That's what they would say.

2449
02:20:14.394 --> 02:20:14.755
Right.

2450
02:20:14.895 --> 02:20:17.237
And I've heard your rebuttal, but I just wanted to express... No, no, I get that.

2451
02:20:17.377 --> 02:20:18.619
I think that's their sentiment.

2452
02:20:18.639 --> 02:20:19.600
I'm not defending them or anything.

2453
02:20:19.660 --> 02:20:20.060
No, I get it.

2454
02:20:20.180 --> 02:20:26.327
I'm trying to provide enough, like, balance to see if we can dance between these two perspectives.

2455
02:20:27.907 --> 02:20:31.369
And a lot of people would say that there's going to be a bloodbath.

2456
02:20:31.750 --> 02:20:31.870
Yeah.

2457
02:20:31.890 --> 02:20:35.512
Because they can't all win big in the way that they're kind of describing.

2458
02:20:35.572 --> 02:20:36.693
So someone's going to have to lose.

2459
02:20:36.773 --> 02:20:42.417
And when one of these players starts to lose big, I think it could, as you say, there could be some kind of domino effect or contraction.

2460
02:20:42.577 --> 02:20:42.797
Yeah.

2461
02:20:42.857 --> 02:20:47.380
And I think the thing that people want to believe is the dot-com bubble thing.

2462
02:20:47.400 --> 02:20:48.781
It's like, it worked out afterwards.

2463
02:20:49.062 --> 02:20:52.804
Because Amazon, Oracle, they didn't die after the dot-com bubble.

2464
02:20:52.824 --> 02:20:53.605
They were actually fine.

2465
02:20:53.625 --> 02:20:54.045
Yeah.

2466
02:20:54.265 --> 02:21:14.387
isn't like that they're bigger companies they're have bigger promises and even i'm not like oracle i actually think could die i r.i.p larry oh it couldn't happen to a nastier man they'll probably like these people do you no i actually why no again i ask this question purely because i want an answer not because i agree or disagree but um why don't you like these these people

2467
02:21:15.027 --> 02:21:19.670
I don't like being misled and I don't think regular people are being misled either.

2468
02:21:20.011 --> 02:21:25.234
And I really don't think that the average person can get away with bullshitting as much of these companies do.

2469
02:21:25.514 --> 02:21:31.658
And I don't think the average person gets anywhere near the level of affordance for failure and lying as these companies do.

2470
02:21:31.938 --> 02:21:40.224
And I think there is a real economic and human cost to allowing these companies to run rampant and promise the world and never really get called up on it.

2471
02:21:40.464 --> 02:22:07.929
the tepid nature of criticism these days is so frustrating there are some really great critics out there the really great people but it's like seeing these ultra rich ultra wealthy ultra powerful people lie through their fucking teeth or misstate or whatever people want to call it it turns my stomach and i hate seeing people being misled and i feel like i write at such length because i really want people to see why i've come to a conclusion am i right am i wrong i think i am of course i do

2472
02:22:09.317 --> 02:22:12.759
But I also, I just find it loathsome.

2473
02:22:12.859 --> 02:22:15.120
I find these companies don't make good products anymore.

2474
02:22:15.340 --> 02:22:16.760
They don't care about their customers.

2475
02:22:18.321 --> 02:22:20.002
And they treat their customers with contempt.

2476
02:22:22.236 --> 02:22:25.539
If people want to go read more about your work, you have a great Substack.

2477
02:22:25.860 --> 02:22:26.440
Ghost, actually.

2478
02:22:26.620 --> 02:22:29.022
It looks exactly like I'd moved off of Substack in 2024.

2479
02:22:29.103 --> 02:22:29.743
Oh, okay.

2480
02:22:30.244 --> 02:22:31.805
And you also have a podcast, you do?

2481
02:22:32.045 --> 02:22:32.686
Yeah, Better Off Lime.

2482
02:22:33.267 --> 02:22:34.568
I'm going to link both of them below.

2483
02:22:34.608 --> 02:22:38.952
So if anyone wants to read more, get more detail and follow Ed, I think it's, I would highly recommend.

2484
02:22:39.132 --> 02:22:40.073
It is fascinating.

2485
02:22:40.093 --> 02:22:40.593
And do you know what?

2486
02:22:41.374 --> 02:22:46.097
One of the things people sometimes struggle with when they listen to podcasts is you get lots of different opinions.

2487
02:22:46.558 --> 02:22:47.738
And weirdly, I think they think of...

2488
02:22:47.899 --> 02:22:52.842
Some people assume podcasts are going to be like one person saying the same thing as the next person and the next person.

2489
02:22:52.902 --> 02:22:53.402
Yeah, yeah, yeah.

2490
02:22:53.462 --> 02:22:56.484
That is just not the nature of information in the world and opinions and progress and discussion.

2491
02:22:57.625 --> 02:22:59.346
What happens is people have different opinions.

2492
02:22:59.406 --> 02:23:06.811
And I think my job, but also the listener's job, is to try and pass through it and over time collect more of these reference points from different people

2493
02:23:08.192 --> 02:23:09.294
and do your own research.

2494
02:23:09.434 --> 02:23:09.654
Yeah.

2495
02:23:09.894 --> 02:23:14.199
Whether it's on your health or whether it's on something like this, is to watch and do your own research and to learn.

2496
02:23:14.900 --> 02:23:20.586
And I would say also, never believe one person, never believe one particular perspective religiously.

2497
02:23:20.906 --> 02:23:24.370
You know, collect a body of evidence and follow the evidence yourself.

2498
02:23:24.730 --> 02:23:29.936
But I love watching your YouTube because it provides a different opinion.

2499
02:23:30.536 --> 02:23:36.419
And that challenges me to think beyond my current opinion about what might be possible.

2500
02:23:36.779 --> 02:23:47.685
So when I've heard you talking about how this is an economic bubble, and I've heard you talk about the CapEx spend with these big sort of frontier AI labs, it really did make me pause for a second.

2501
02:23:47.745 --> 02:23:53.228
And it really did make me consider that there could be a bit of figazi going on here.

2502
02:23:53.768 --> 02:23:57.572
And then it made me reflect on history and go, you know, through history, there's always a bit of figase in these moments.

2503
02:23:57.632 --> 02:23:59.393
And, oh, that's an interesting take.

2504
02:23:59.413 --> 02:24:02.656
And what's going to happen in 2027, 2028 when there's a bit of a market pullback?

2505
02:24:02.676 --> 02:24:05.038
And so I highly recommend people go watch because you do.

2506
02:24:05.058 --> 02:24:06.320
You challenge me to think differently.

2507
02:24:06.340 --> 02:24:06.900
Yeah.

2508
02:24:07.501 --> 02:24:09.242
And we need some of those contrarian voices to...

2509
02:24:10.303 --> 02:24:11.264
to have honest discussions.

2510
02:24:11.644 --> 02:24:12.545
So thank you for doing what you do.

2511
02:24:12.945 --> 02:24:13.606
Really appreciate it.

2512
02:24:13.626 --> 02:24:16.108
And I find you to be a very compelling, captivating communicator.

2513
02:24:16.368 --> 02:24:17.849
And I feel like I've learned a lot today.

2514
02:24:18.490 --> 02:24:19.150
So I appreciate that.

2515
02:24:19.190 --> 02:24:20.131
We have a closing tradition.

2516
02:24:20.491 --> 02:24:20.692
Yeah.

2517
02:24:21.052 --> 02:24:23.854
Where the last guest leaves a question for the next guest, not knowing who they're leaving it for.

2518
02:24:24.435 --> 02:24:33.962
And the question left for you is, given that high quality relationships are important for health and longevity, what should we be doing to improve our relationships and social connection?

2519
02:24:34.883 --> 02:24:38.344
So this is actually connected to the AI bubble.

2520
02:24:38.684 --> 02:24:40.005
So I'm a critic.

2521
02:24:40.045 --> 02:24:41.605
I'm a skeptic.

2522
02:24:42.426 --> 02:24:52.129
I have found that showing and appreciating and loving the people around you and uplifting them and raising them up as you succeed is the way we do that.

2523
02:24:52.249 --> 02:24:53.570
Your success should be everyone around you.

2524
02:24:53.590 --> 02:24:54.190
It's not economic.

2525
02:24:54.290 --> 02:24:57.331
It's talking about Matt Hughes for a while.

2526
02:24:57.411 --> 02:24:58.251
Made me really happy.

2527
02:24:58.651 --> 02:25:03.033
This whole thing has been at times quite grueling and quite negative and quite brutal.

2528
02:25:04.038 --> 02:25:19.168
But the love I've found and the joy I've found from community and the people around... Because even in the small groups of haters, even like Gary Marcus and so the people I talked to, Edward Ngueso Jr., Molly White, Brian Merchant, there are so many people who have been loving and caring.

2529
02:25:19.188 --> 02:25:29.255
And I think within especially these very critical moments, when you're very much dialing in on how negative things are, how bad things are, finding the people who...

2530
02:25:30.155 --> 02:25:31.396
Maybe find it repulsive, too.

2531
02:25:31.457 --> 02:25:33.639
Finding the people... Finding your people.

2532
02:25:33.679 --> 02:25:34.099
Who can be...

2533
02:25:34.660 --> 02:25:35.821
The people who will talk to you about it.

2534
02:25:35.901 --> 02:25:39.385
Even, like, Troy and Jake, my trainers, who are so excited about this.

2535
02:25:40.526 --> 02:25:42.508
Even talking to them about this shit as normal people.

2536
02:25:43.048 --> 02:25:45.671
Knowing that there are people there going through their own struggles, but also...

2537
02:25:46.712 --> 02:26:11.836
to just give you the perspective and also remind you that you are human too and focus i know this is kind of a all over the place point but it's just it's really easy to get hard locked on everything in life and to kind of get away from why you do things and focus too much on the work when the most important thing at times is just to know there are other people feeling the way you do and when i hear from my listeners my readers a lot the most common thing i feel is they feel like they have a voice and they feel like someone is there for you

2538
02:26:12.476 --> 02:26:18.161
And I don't think it can be understated how much it means when you just reach out to someone you love and tell them you love them.

2539
02:26:18.201 --> 02:26:19.242
Tell them their shit rocks.

2540
02:26:19.642 --> 02:26:20.743
Say that their shit bangs.

2541
02:26:21.064 --> 02:26:26.388
Tell everyone when you like an artist or a writer's thing or a podcast like this, tell them you fucking love it.

2542
02:26:26.708 --> 02:26:27.950
We don't do this enough.

2543
02:26:28.570 --> 02:26:29.551
And we need to do it more.

2544
02:26:30.212 --> 02:26:31.433
Well, that's a good closing message.

2545
02:26:31.713 --> 02:26:35.916
So if you have enjoyed the conversation today with Ed, please do let Ed know that you love it down below.

2546
02:26:36.977 --> 02:26:39.299
But please do leave your opinions down below and I shall read all of them.

2547
02:26:39.799 --> 02:26:40.600
Ed, thank you so much.

2548
02:26:40.640 --> 02:26:44.603
I'll link to your website, but also to your YouTube channel where people can learn more.

2549
02:26:44.623 --> 02:26:47.365
And I would highly recommend you do because it is truly fascinating.

2550
02:26:47.405 --> 02:26:52.009
And I think we need more voices that are demystifying a lot of the figazi and the narrative in this moment in time.

2551
02:26:52.309 --> 02:26:53.070
And you're certainly one of them.

2552
02:26:53.110 --> 02:26:54.010
I really enjoyed the conversation.

2553
02:26:54.190 --> 02:26:54.691
Thank you so much.