WEBVTT

1
00:00:08.114 --> 00:00:14.717
Hi, and welcome to Technically Creative's Week in Review, brought to you by Kubrick Labs, the AI and IT partner for creative companies.

2
00:00:15.358 --> 00:00:21.301
Every week, I spend about 20 minutes walking you through the most important news from the increasingly tangled worlds of tech and creativity.

3
00:00:21.501 --> 00:00:28.204
This week, the man who built the chips that run the AI boom resigned, four days after his board said it had full confidence in him.

4
00:00:29.365 --> 00:00:35.028
Disney, which once called an AI company a bottomless pit of plagiarism, named a new official AI partner.

5
00:00:35.762 --> 00:00:42.213
Microsoft's email went down in 74 countries, and Microsoft felt it needed to tell us that AI agents didn't do it.

6
00:00:42.794 --> 00:00:45.438
I don't know why, but that makes me think that AI agents did it.

7
00:00:46.267 --> 00:00:52.349
An AI singer won three awards voted on by the AI music community, which sounds like a snake eating its own tail.

8
00:00:52.549 --> 00:00:55.910
A journalist spent hours trying to teach a robo duck to moonwalk.

9
00:00:56.370 --> 00:00:59.371
And on Friday afternoon, the United States stopped flying altogether.

10
00:00:59.411 --> 00:01:03.973
And what's particularly annoying is I was on one of those flights, and I was in it for 20 hours.

11
00:01:04.173 --> 00:01:06.193
So yeah, this week was cuckoo, a little bit scary.

12
00:01:06.713 --> 00:01:10.995
And I always wonder, do you think there was a prequel to Terminator that was funny?

13
00:01:11.675 --> 00:01:13.796
Because that's what I think that we're living through right now.

14
00:01:13.977 --> 00:01:16.418
The absurdist first steps of Skynet.

15
00:01:17.019 --> 00:01:20.581
I think if I were writing this as a novel, the publisher would say, pick a genre.

16
00:01:21.582 --> 00:01:23.003
But anyway, let's start with the news.

17
00:01:23.383 --> 00:01:24.864
Jensen Huang leaves NVIDIA.

18
00:01:24.984 --> 00:01:33.269
On Thursday, NVIDIA announced that Jensen Huang, the co-founder who has run the company for 33 years, has resigned as chief executive and left the board effective immediately.

19
00:01:33.710 --> 00:01:38.573
If you're not a chip person, NVIDIA makes the processors that almost every AI model is trained and run on.

20
00:01:38.733 --> 00:01:40.494
And listen, you don't need much update on this.

21
00:01:40.694 --> 00:01:42.615
It's been pretty heavily reported everywhere.

22
00:01:42.755 --> 00:01:44.316
I think he's going to come back personally.

23
00:01:44.916 --> 00:01:47.178
And there are loads of great people in that company.

24
00:01:47.238 --> 00:01:49.139
It's been running for 33 years.

25
00:01:49.599 --> 00:01:52.621
It's not like Anthropic, which is only five years old.

26
00:01:52.821 --> 00:01:58.404
This is a company with a lot of really good, smart people at the top who can all take that role.

27
00:01:58.464 --> 00:02:01.346
There's a petition right now from a lot of the employees to bring him back.

28
00:02:01.386 --> 00:02:03.407
I think the stock market wants him back.

29
00:02:03.447 --> 00:02:05.188
I think all of the shareholders want him back.

30
00:02:05.368 --> 00:02:10.832
There are big questions around China and why they were selling to China, but let's be real.

31
00:02:11.292 --> 00:02:14.415
$4.6 billion is why they were selling to China.

32
00:02:14.595 --> 00:02:19.818
Sure, American companies need to be aware that their technology is going to be used by foreign powers.

33
00:02:20.019 --> 00:02:24.222
But at the end of the day, Jensen Wang and the board are solely focused on one thing.

34
00:02:24.922 --> 00:02:26.463
making the stock's price go up.

35
00:02:26.823 --> 00:02:28.224
They're not a military company.

36
00:02:28.304 --> 00:02:30.985
Sure, their tools have military applications.

37
00:02:31.126 --> 00:02:33.307
So yeah, I think there are big questions to be asked.

38
00:02:33.347 --> 00:02:38.530
I think we should probably expect some sort of sanctions or new limitations on the sale of chips to China.

39
00:02:38.750 --> 00:02:40.911
But Wang's down, he's not necessarily out.

40
00:02:41.031 --> 00:02:44.752
And then in other news, Disney found an AI it likes, one that it controls.

41
00:02:45.352 --> 00:02:50.774
On Monday at Adobe's annual Max Conference, Disney named Adobe its official AI innovation partner.

42
00:02:51.194 --> 00:02:56.896
Disney will use Adobe's Firefly Foundry, a service that builds private AI models for individual companies.

43
00:02:57.316 --> 00:03:01.778
And it's gonna make models trained on Disney's own characters, environments, and visual styles.

44
00:03:02.378 --> 00:03:14.942
Walt Disney Imagineering, the team that designs the theme parks, is already using one that turns hand-drawn sketches into rendered concept art, and another that generates on-model assets for Mickey and Friends, Frozen, Moana, and Cars.

45
00:03:15.622 --> 00:03:19.983
In June last year, Disney sued Mint Journey and called it a bottomless pit of plagiarism.

46
00:03:20.083 --> 00:03:22.964
In December, it put a billion dollars into OpenAI.

47
00:03:23.364 --> 00:03:29.366
Now it has an official AI partner, and that partner gets to promote itself at the Walt Disney World and on Disney Cruise Lines.

48
00:03:29.806 --> 00:03:31.327
So Disney doesn't hate AI.

49
00:03:31.627 --> 00:03:37.172
Disney hates AI it doesn't own, which, to be fair, is exactly how Disney feels about just about everything.

50
00:03:37.932 --> 00:03:46.679
And it's probably the template for every studio, agency, publisher, or anyone sitting on valuable IP, a private model trained on your own stuff that never leaves the building.

51
00:03:47.485 --> 00:03:55.512
At the same conference, Adobe announced that its creative agent now runs across Photoshop, Premiere, Illustrator, and After Effects, and previewed something more interesting.

52
00:03:56.033 --> 00:04:06.422
Next year, in a private beta, you'll be able to point the AI assistant of your choice at your Photoshop or Premiere files and have it make changes locally, with your layers and timelines intact.

53
00:04:06.962 --> 00:04:09.545
Adobe spent 40 years teaching people Photoshop.

54
00:04:10.185 --> 00:04:13.950
It's now teaching Photoshop to take instructions from somebody else's software.

55
00:04:14.371 --> 00:04:14.972
The good news?

56
00:04:15.292 --> 00:04:16.814
The robot can retouch the shot.

57
00:04:16.914 --> 00:04:20.960
The bad news is it still can't tell you to feel good about the client wanting the logo to be bigger.

58
00:04:21.208 --> 00:04:23.449
David Ellison wants AI to make movies cheaper.

59
00:04:23.949 --> 00:04:34.575
On Paramount Skydance's first earnings call since it took over Warner Brothers, chief executive David Ellison called AI potentially the biggest creative unlock in modern history, bigger than CGI.

60
00:04:35.135 --> 00:04:39.217
His example was The Terminator, which cost about $4 million in 1984.

61
00:04:39.818 --> 00:04:43.519
Remake it shot for shot today, he said, and it would cost 60 to 70 million.

62
00:04:44.680 --> 00:04:49.344
His argument is that AI brings costs down so that studios can take risks on original films.

63
00:04:49.984 --> 00:04:53.607
That's the right question, and it's the counterfactual I always want to ask.

64
00:04:53.967 --> 00:04:57.870
Does cheaper production mean more films or the same films with fewer people?

65
00:04:58.511 --> 00:05:00.192
Hollywood's answer is usually the second one—

66
00:05:00.392 --> 00:05:05.856
But I think we got to give David Ellison some credit here because he seems to be talking about the first one, and that's the one we like.

67
00:05:05.916 --> 00:05:08.117
And then there's some other news in the Skydance world.

68
00:05:08.177 --> 00:05:19.825
James Gunn and Peter Safran are stepping down as the heads of DC Studios, the Warner Brothers division that makes Superman and Batman, and Jeremy Latcham, a longtime Marvel executive, takes over on December 1st.

69
00:05:20.305 --> 00:05:24.428
Gunn and Safran will keep making their own films, including the next Superman, Man of Tomorrow.

70
00:05:24.908 --> 00:05:27.210
And as a comic book fan, I have to say...

71
00:05:28.195 --> 00:05:29.095
I'm bummed about this.

72
00:05:29.295 --> 00:05:34.537
I know that the box office for DC didn't live up to a lot of the expectations that people had of James Gunn.

73
00:05:34.797 --> 00:05:41.639
But what he and Peter Safran brought to the table at DC was something that is hard to find nowadays in superhero movies.

74
00:05:41.819 --> 00:05:43.199
Real risk and surprise.

75
00:05:43.339 --> 00:05:47.181
There have been a lot of Superman movies, but that one was pretty unique.

76
00:05:47.421 --> 00:05:49.001
And I think it was a fascinating vision.

77
00:05:49.601 --> 00:05:51.622
Lanterns took a completely different turn.

78
00:05:52.944 --> 00:05:56.788
I think it was something that was starting to come together, and I would have liked to have seen where it went.

79
00:05:56.868 --> 00:06:01.193
And sure, DC has always been the darker, younger brother of Marvel.

80
00:06:01.233 --> 00:06:04.576
But James and Peter found a way to make it lighter and more interesting.

81
00:06:04.656 --> 00:06:12.745
And in a world where everybody's saying that we need more risks and more creativity, I think that that was not lacking in what we were seeing coming out of DC.

82
00:06:13.125 --> 00:06:15.226
I wish Jeremy all the luck in the world.

83
00:06:15.466 --> 00:06:18.567
At the end of the day, I'm rooting for superhero films.

84
00:06:18.807 --> 00:06:20.767
I am a comic book nut.

85
00:06:20.947 --> 00:06:22.008
I love DC.

86
00:06:22.108 --> 00:06:22.968
I love Vertigo.

87
00:06:23.088 --> 00:06:30.250
I just am happy I get to live in a world where somebody thinks it's good business to make a $200 million superhero movie.

88
00:06:30.790 --> 00:06:31.570
I think that's great.

89
00:06:31.650 --> 00:06:37.112
And I thought the last Superman movie was a celebration of a love for

90
00:06:38.178 --> 00:06:40.420
a character that can be a little bit one note sometimes.

91
00:06:40.720 --> 00:06:42.581
So anyway, sad to see that happen.

92
00:06:43.202 --> 00:06:47.005
I don't think that, you know, David Ellison and Skydance are evil for, for doing it.

93
00:06:47.425 --> 00:06:51.268
Um, and at the end of the day, I hope they take what was good about it.

94
00:06:51.388 --> 00:06:52.969
And I hope they keep taking big swings.

95
00:06:53.029 --> 00:06:53.710
Cause, you know,

96
00:06:54.153 --> 00:06:56.056
I really appreciate that that's what DC was doing.

97
00:06:56.196 --> 00:06:59.460
And then in some positive news about AI and art, here's a lovely one.

98
00:06:59.860 --> 00:07:12.457
The art newspaper reports that conservators were storing a comfortable corner, a painting from around 1888 by the American impressionist William Merritt Chase, used AI to compare a black and white photograph taken before...

99
00:07:12.917 --> 00:07:15.720
a 1960s restoration with the painting as it is now.

100
00:07:16.041 --> 00:07:18.183
The software showed where the two had diverged.

101
00:07:18.664 --> 00:07:23.329
A lost curl of hair, a faded decorative screen, passages that had been painted over.

102
00:07:23.870 --> 00:07:27.354
It also helped confirm that Chase's signature had faded almost to nothing.

103
00:07:27.774 --> 00:07:33.241
And the conservator, Susan Ciano, thinks an earlier restorer may have mistaken it for part of the carpet.

104
00:07:33.441 --> 00:07:39.006
Siano says the AI was helpful largely because of its speed, and her team didn't reinstate everything it flagged.

105
00:07:39.446 --> 00:07:43.530
The painting also goes back on the show at the Parrish Art Museum on Long Island this month.

106
00:07:44.130 --> 00:07:51.457
After years of AI imitating dead painters, it's found a job everyone can agree on, noticing that a previous restorer painted over the artist's name.

107
00:07:51.677 --> 00:07:53.858
And then AI Music got its own awards show.

108
00:07:53.978 --> 00:08:09.906
On Friday night at the Mayan in downtown Los Angeles, the first AI Music Awards handed AI Artist Song and Album of the Year to Zanya Monet, an AI-generated singer created by the songwriter Talisha Nikki Jones, who writes the lyrics and uses AI to make the music.

109
00:08:10.387 --> 00:08:17.250
According to Reuters, the winners were chosen by AI artists who are members of the Sonic Intelligence Academy, the company that runs the awards.

110
00:08:17.770 --> 00:08:20.632
The AI music charts and the tools that detect AI music.

111
00:08:20.992 --> 00:08:27.618
There's a serious debate underneath it about where the artistry sits, and Jones writing every word is a different thing from a fake Beyonce.

112
00:08:28.058 --> 00:08:30.961
But my favorite line of the night came from the producer Drake Murphy.

113
00:08:31.461 --> 00:08:35.965
He said that in a world where everybody calls AI fake, the human behind it is real.

114
00:08:36.305 --> 00:08:39.508
He said that while presenting the award for best AI hip-hop song.

115
00:08:39.728 --> 00:08:41.190
I don't know what we're talking about here.

116
00:08:41.913 --> 00:08:43.494
And then finally, there was a duck.

117
00:08:43.914 --> 00:08:57.138
Will Nutt, Wired's senior AI writer, spent part of his week using OpenAI's newest model to train a virtual version of the MicroDuck, a 10-inch walking robot from Hugging Face's robotics team, to find his Bengal cat, Leona, and give her treats.

118
00:08:57.718 --> 00:09:00.619
He also fed it a GIF of Michael Jackson and asked it to moonwalk.

119
00:09:01.079 --> 00:09:05.244
After hours of effort, he got what he described as a very jaunty backward jog.

120
00:09:05.704 --> 00:09:08.047
The physical duck hasn't delivered a single treat yet.

121
00:09:08.867 --> 00:09:12.832
Trillions of dollars of AI investment, and we've arrived at a feline room service.

122
00:09:12.932 --> 00:09:15.535
Honestly, that may be the best news of the week.

123
00:09:16.215 --> 00:09:16.996
So that's the news.

124
00:09:17.156 --> 00:09:25.223
And I want to do a deep dive into something that I had personally to live through as I was stuck in a plane for 20 hours on Friday.

125
00:09:25.383 --> 00:09:31.068
And it got me wondering about all this new technology we're using, but what the infrastructure is we're using it on.

126
00:09:31.428 --> 00:09:34.510
Are we using tomorrow's technology on yesterday's infrastructure?

127
00:09:34.530 --> 00:09:35.131
Yeah.

128
00:09:35.191 --> 00:09:36.612
I think we're starting to see that we are.

129
00:09:37.332 --> 00:09:48.078
At 3.54 on Friday afternoon Eastern Time, the Federal Aviation Administration grounded every departing flight in the United States after communication problems at several of the big regional centers.

130
00:09:48.198 --> 00:09:49.879
It lasted more than four hours.

131
00:09:50.059 --> 00:09:55.082
About 600 flights were diverted, and it's only the second nationwide ground stop since 9-11.

132
00:09:55.202 --> 00:10:01.085
Transportation Secretary and MTV's Real World alum, Sean Duffy, blamed old legacy equipment—

133
00:10:01.545 --> 00:10:05.227
Routers overloaded with data that started giving inconsistent readings.

134
00:10:05.787 --> 00:10:14.091
The Homeland Security Secretary said that they were looking at incompatibilities between that old equipment and the new technology that was being installed just a couple days before.

135
00:10:14.831 --> 00:10:17.253
There's no sign of a cyber attack or anything like that.

136
00:10:17.713 --> 00:10:21.235
CNN reported that some air traffic controllers don't buy the explanation.

137
00:10:21.555 --> 00:10:24.356
And Duffy himself says they're still working out what went wrong.

138
00:10:24.436 --> 00:10:26.217
Now, here's where the timing is suspicious.

139
00:10:26.697 --> 00:10:30.399
Three days earlier, on Tuesday, the FAA expanded SMART—

140
00:10:30.759 --> 00:10:35.262
It's AI-assisted traffic planning system to all 22 of those centers.

141
00:10:35.882 --> 00:10:40.505
Duffy says smart had nothing to do with Friday and only the old equipment failed.

142
00:10:40.765 --> 00:10:44.327
And I believe him, but I think that that's only one way of looking at the problem.

143
00:10:44.527 --> 00:10:49.651
New technologies require bigger and faster data infrastructures to support them.

144
00:10:50.671 --> 00:10:52.232
So it may not be the AI's fault.

145
00:10:52.352 --> 00:10:54.193
It may have operated exactly as planned.

146
00:10:54.654 --> 00:10:56.515
A whole bunch of agents didn't go rogue.

147
00:10:56.815 --> 00:10:59.977
But that doesn't mean it didn't require a lot more energy and data

148
00:11:00.994 --> 00:11:05.278
that then left the old system wanting for what it needed to operate.

149
00:11:07.419 --> 00:11:09.601
Smart first launched on September 21st.

150
00:11:09.741 --> 00:11:16.187
That same morning, an old Verizon circuit that served the radar facility that handles Newark and Philadelphia failed.

151
00:11:16.687 --> 00:11:20.050
And when controllers switched to the backup, they found contractors had cut it.

152
00:11:20.631 --> 00:11:24.494
The FAA's administrator, Brian Bedford, put it perfectly that day.

153
00:11:24.914 --> 00:11:27.377
We're pulling copper up and putting fiber down.

154
00:11:28.017 --> 00:11:31.360
But that's the same data architecture that we've had for 50 years.

155
00:11:31.900 --> 00:11:37.304
So that means that twice now, the AI has turned up for its first day and the building has fallen down around it.

156
00:11:37.385 --> 00:11:41.688
And on the same week, on Thursday, just the day before, Microsoft 365 went down.

157
00:11:42.389 --> 00:11:46.652
Outlook teams co-pilot with outage reports from 74 countries within an hour.

158
00:11:47.272 --> 00:11:49.814
Governments, courts, hospitals, airlines, and banks.

159
00:11:50.235 --> 00:11:53.698
Twelve hours in, Microsoft said it had marginal success fixing it.

160
00:11:54.258 --> 00:11:56.079
Full service came back on Saturday.

161
00:11:56.439 --> 00:11:59.401
The cause was a network change during a capacity upgrade.

162
00:11:59.821 --> 00:12:07.306
And Microsoft told everyone it was not related to cyber or to AI agents, which, when they say that, makes me think it probably was.

163
00:12:07.786 --> 00:12:13.090
And again, this doesn't sound like something where AI went rogue, but hear the wording.

164
00:12:13.170 --> 00:12:16.011
They said it was a quote-unquote capacity upgrade.

165
00:12:17.252 --> 00:12:20.074
All these new tools require a lot more capacity.

166
00:12:20.274 --> 00:12:23.216
It's what we were talking last week about with data centers.

167
00:12:25.277 --> 00:12:26.600
Our grid needs updating.

168
00:12:26.760 --> 00:12:28.364
Our data systems need updating.

169
00:12:28.424 --> 00:12:33.735
And so when I line the week up, nothing broke at the smart bit, at the AI bit.

170
00:12:34.096 --> 00:12:35.659
And it wasn't AI running amok.

171
00:12:36.425 --> 00:12:38.447
The AI traffic system did its job.

172
00:12:38.807 --> 00:12:40.869
It's the routers underneath it that didn't.

173
00:12:41.450 --> 00:12:43.672
Microsoft's AI assistant didn't go rogue.

174
00:12:43.732 --> 00:12:46.094
Somebody changed a network setting during an upgrade.

175
00:12:46.634 --> 00:12:53.861
And NVIDIA, the company that designs the most sophisticated computing hardware on the planet, didn't lose its chief executive to a technical problem.

176
00:12:54.221 --> 00:12:58.545
Every failure this week happened at the oldest, most ordinary layer of the system.

177
00:12:58.585 --> 00:13:01.808
And the oldest, most ordinary layer of the system is us.

178
00:13:02.769 --> 00:13:05.992
People are the messiest, most fragile part of any technology.

179
00:13:06.432 --> 00:13:07.633
We reuse passwords.

180
00:13:07.653 --> 00:13:08.454
We click the link.

181
00:13:08.814 --> 00:13:10.355
We take the work laptop home.

182
00:13:10.395 --> 00:13:12.157
We trust the nice person on the phone.

183
00:13:12.597 --> 00:13:13.578
We post our runs.

184
00:13:13.958 --> 00:13:17.201
An engineer can rip out a 50-year-old router and put in a new one.

185
00:13:17.781 --> 00:13:19.983
Nobody's found a way to patch a human being.

186
00:13:20.383 --> 00:13:25.928
And the more sophisticated the machinery gets, the more weak points move to the person standing next to it.

187
00:13:27.442 --> 00:13:31.606
Just yesterday, the LA Times showed us what that looks like at its most serious.

188
00:13:32.146 --> 00:13:36.831
It reported on how Iran picked its targets when it struck American bases in the Middle East.

189
00:13:37.031 --> 00:13:38.933
It didn't hack military systems.

190
00:13:39.493 --> 00:13:46.400
It used data from service members' social media, fitness apps, and even ads to work out patterns of life on the bases.

191
00:13:46.940 --> 00:13:48.101
Who goes where and when?

192
00:13:48.121 --> 00:13:49.502
This is commercial data.

193
00:13:50.323 --> 00:13:55.966
And then when they paired it with satellite imagery reportedly from Russia, that built a surprisingly accurate picture.

194
00:13:56.546 --> 00:14:05.410
A Sky News investigation over the summer found that more than 1,300 people logging workouts on Stratva, a fitness tracking ad, on U.S. bases in the region.

195
00:14:06.270 --> 00:14:07.491
People died in those attacks.

196
00:14:07.571 --> 00:14:08.592
This isn't a joke.

197
00:14:08.812 --> 00:14:16.199
The joke, if there is one, is the decade we spent as individuals saying, I've got nothing to hide, who cares what they know about me?

198
00:14:16.399 --> 00:14:19.963
Turns out the answer is, even your morning run is potentially important.

199
00:14:20.543 --> 00:14:30.652
And in the same pattern but with a different target, on October 29th, the crypto exchange Binance lost an estimated $6.3 billion, with about 3 billion cents frozen.

200
00:14:31.353 --> 00:14:37.718
Binance's primary finding was that it may have just started with a compromised laptop belonging to one DevOps engineer.

201
00:14:39.499 --> 00:14:43.922
$6.3 billion because one guy's laptop was compromised.

202
00:14:46.605 --> 00:14:48.746
And so, yeah, we have old infrastructure.

203
00:14:49.347 --> 00:14:51.368
We have human beings that are fallible.

204
00:14:53.049 --> 00:14:57.653
But now we have this AI that can do everything all at once.

205
00:14:59.430 --> 00:15:04.875
It used to be that somebody like Jensen Huang, everybody knows he's important.

206
00:15:05.115 --> 00:15:10.360
The random guys on the bases logging their workouts, not everybody's sure how important they are.

207
00:15:10.941 --> 00:15:15.925
But AI can be assessing all weaknesses in a system all at the same time.

208
00:15:16.005 --> 00:15:22.952
And so we have to get better at updating not just the physical systems, but our own relationship to our own data in the technological world.

209
00:15:24.089 --> 00:15:26.191
We have to put it behind stronger firewalls.

210
00:15:26.271 --> 00:15:30.415
Because science fiction's always told us that the machine would turn on us.

211
00:15:31.116 --> 00:15:33.198
What's actually happening is a lot smaller.

212
00:15:33.698 --> 00:15:35.680
The machine is getting very good at reading us.

213
00:15:35.901 --> 00:15:39.604
Finding the person with the key used to be a slow, expensive human work.

214
00:15:40.065 --> 00:15:48.072
Going through LinkedIn, mapping who works where, who has which login, who posts their gym sessions, who's just started a new job and might fall for a fake interview.

215
00:15:48.092 --> 00:15:48.673
Yeah.

216
00:15:48.773 --> 00:15:53.396
That's precisely the kind of patient, tedious research AI agents are brilliant at.

217
00:15:53.956 --> 00:15:55.577
So AI is doing two things at once.

218
00:15:55.877 --> 00:15:59.860
It's finding our weak spots, and it's finding out what's valuable about us.

219
00:16:00.440 --> 00:16:03.062
What we know, what we can open, who we sit next to.

220
00:16:03.462 --> 00:16:05.243
I think that's going to get dramatically cheaper.

221
00:16:05.623 --> 00:16:09.706
In the AI era, you don't have to be an important person to possess important information.

222
00:16:10.006 --> 00:16:14.008
You just have to be ignorant as to how important the information you touch is.

223
00:16:15.869 --> 00:16:18.051
And that has to be useful to someone.

224
00:16:18.872 --> 00:16:21.994
And that's why I keep feeling that every one of these stories now touches everything.

225
00:16:22.575 --> 00:16:24.596
A running app becomes military intelligence.

226
00:16:24.777 --> 00:16:27.919
One engineer's laptop becomes a $6 billion backdoor.

227
00:16:28.480 --> 00:16:31.602
An email outage closes courts on the other side of the world.

228
00:16:32.222 --> 00:16:37.587
There isn't a separate tech bit of life anymore, sitting off to one side where the nerds worry about it.

229
00:16:37.667 --> 00:16:39.668
It's all one system and we're all a part of it.

230
00:16:40.349 --> 00:16:43.932
And for a really long time, creative people and creative companies have felt like

231
00:16:44.960 --> 00:16:46.861
the least important part of that process.

232
00:16:46.921 --> 00:16:51.003
Creative people tend to think, that's defense contractors and crypto exchanges.

233
00:16:51.424 --> 00:16:52.925
Who on earth wants our stuff?

234
00:16:53.425 --> 00:16:54.245
How important is it?

235
00:16:54.385 --> 00:16:55.166
It's a pitch deck.

236
00:16:55.306 --> 00:16:56.086
It's a rough cut.

237
00:16:56.407 --> 00:16:57.747
It's a song that isn't finished.

238
00:16:58.708 --> 00:17:00.349
Honestly, a lot of people feel that way.

239
00:17:01.191 --> 00:17:04.772
But governments have understood for a very long time that stories are power.

240
00:17:04.892 --> 00:17:12.115
In 2014, the U.S. government said North Korea was behind the attack of Sony Pictures over a comedy, The Interview, about topping off Kim Jong-un.

241
00:17:12.715 --> 00:17:17.117
Unreleased films, scripts, and executive emails and people's salaries all ended up online.

242
00:17:17.677 --> 00:17:21.239
A nation state attacked a film studio because of a Seth Rogen movie.

243
00:17:21.379 --> 00:17:22.740
And I'd go one step further.

244
00:17:23.160 --> 00:17:27.082
I think one of the reasons the West has led the world for the last century is its creativity.

245
00:17:27.563 --> 00:17:36.288
The films, the music, the advertising, the television, the stories, that traveled everywhere and made people around the world want what we had and believe what we believed.

246
00:17:36.888 --> 00:17:38.269
You can call it soft power.

247
00:17:38.289 --> 00:17:39.470
You can call it propaganda.

248
00:17:39.911 --> 00:17:41.112
Honestly, it's been both.

249
00:17:41.532 --> 00:17:43.634
And every serious country on earth knows it.

250
00:17:44.255 --> 00:17:50.720
That's why so many of them are pouring money into their own film industries, their own streaming platforms, their own pop stars.

251
00:17:51.441 --> 00:17:53.883
So what we make isn't just commercially valuable.

252
00:17:54.183 --> 00:17:55.645
It's strategically valuable.

253
00:17:55.985 --> 00:17:57.787
And the people who make it are valuable too.

254
00:17:59.146 --> 00:18:00.627
Think about what it means in practice.

255
00:18:00.987 --> 00:18:03.408
Your unreleased script lives on someone's laptop.

256
00:18:03.448 --> 00:18:07.349
Your rushes sit in a storage account that an assistant editor logs into from home.

257
00:18:07.790 --> 00:18:12.552
Your agency's client data lives on a Microsoft account that, as we learned Thursday, isn't always there.

258
00:18:13.312 --> 00:18:15.213
And go back to Disney's deal with Adobe.

259
00:18:15.513 --> 00:18:19.455
A private AI model trained on every character Disney owns is a wonderful tool.

260
00:18:19.935 --> 00:18:24.697
It's also quite possibly one of the most valuable creative files ever assembled in one place.

261
00:18:25.497 --> 00:18:28.319
That's the sort of thing someone somewhere is going to want.

262
00:18:28.359 --> 00:18:32.542
And I guess the point I want to make to people is you don't have to be a defense contractor to be a target.

263
00:18:33.002 --> 00:18:35.083
You just need to work near something valuable.

264
00:18:35.384 --> 00:18:37.825
And in this business, a lot of people do.

265
00:18:37.865 --> 00:18:44.529
If you're working on Nike's summer ad for next year, you know Nike's whole marketing plan ahead of time.

266
00:18:45.590 --> 00:18:49.633
That's incredibly valuable data that would be incredibly valuable to anybody.

267
00:18:50.645 --> 00:18:51.607
So what do we actually do?

268
00:18:51.667 --> 00:18:53.030
I'm not a doomer about all this.

269
00:18:53.350 --> 00:18:55.675
These tools are going to make enormous things possible.

270
00:18:56.035 --> 00:18:59.822
And small creative companies will be able to do things only giant ones used to be able to.

271
00:19:00.582 --> 00:19:03.644
But if the human layer is the weak point, that's where the work starts.

272
00:19:04.184 --> 00:19:06.846
You got to assume people want your stuff because they do.

273
00:19:06.986 --> 00:19:09.328
They might not even know yet why they want it.

274
00:19:09.648 --> 00:19:11.329
They'll figure it out once they see it.

275
00:19:11.950 --> 00:19:17.673
You got to assume the fake interview, the fake supplier, and the fake producer on the phone are coming, and that they'll be convincing.

276
00:19:17.733 --> 00:19:26.219
You got to give AI agents the smallest set of permissions so that they can do their jobs with, and make sure a named human knows exactly what each one can touch and what they can't.

277
00:19:26.339 --> 00:19:32.842
So yes, I guess I'm always just banging on about security, but it needs to be a much bigger part of the way we operate as creative companies.

278
00:19:33.063 --> 00:19:40.907
So yeah, this week, the AI industry lost its most famous executive, Microsoft lost our email, and America lost the ability to get planes off the ground.

279
00:19:41.447 --> 00:19:44.128
None of it was caused by an AI agent, or so they say.

280
00:19:44.809 --> 00:19:46.450
But that kind of proves my point.

281
00:19:46.770 --> 00:19:48.611
The machines aren't the weak spot we are.

282
00:19:48.991 --> 00:19:53.053
And the machines are getting very, very good at finding those weak spots or causing them.

283
00:19:53.856 --> 00:19:54.698
So thanks for listening.

284
00:19:54.799 --> 00:19:57.125
Tomorrow we will be back with another interview episode.

285
00:19:57.686 --> 00:19:59.351
Thank you for listening, and we'll see you next week.
