WEBVTT
1
00:00:00.051 --> 00:00:02.593
Hi, this is the Week in Review from Technically Creative.
2
00:00:02.793 --> 00:00:13.001
I'm Orlando Wood, and over the next 20 minutes, I'm going to give you an overview of what I felt the most important news was last week in the ever-increasingly interdependent worlds of tech and the creative industries.
3
00:00:13.121 --> 00:00:14.843
This is the Week in Review, September 28, 2026.
4
00:00:16.757 --> 00:00:24.103
This week, it seemed like creative leaders started coming out of their dark AI dens and declaring proudly, yeah, we're using AI.
5
00:00:24.183 --> 00:00:24.703
What of it?
6
00:00:24.903 --> 00:00:33.029
It reminds me that an exec at Warner Brothers once said to me, everybody says they hate AI, but everybody's coming back from the bathroom with really great ideas.
7
00:00:33.830 --> 00:00:39.413
And that may be the best description I've heard of the creative industry's relationship with AI over the last few years.
8
00:00:39.793 --> 00:00:41.194
Publicly, horror.
9
00:00:41.554 --> 00:00:44.255
Privately, lots of experimentation and interest.
10
00:00:44.735 --> 00:01:01.404
As someone who helps big companies adopt software that uses AI, for me it's felt a bit like getting to date the prettiest girl in high school, but she refuses to acknowledge you in the hallway, she tells you you can't tell your friends, but she will make out with you under the bleachers on Tuesdays and Wednesdays during study hall.
11
00:01:01.984 --> 00:01:03.785
A24 was one of our first clients.
12
00:01:04.106 --> 00:01:06.187
We're allowed to say we work with them colloquially.
13
00:01:06.427 --> 00:01:08.989
We're just not allowed to say what we actually do for them.
14
00:01:09.529 --> 00:01:17.775
And it's the same with most of our industry clients, which is strange because the companies we work with are incredibly thoughtful about how they use this technology.
15
00:01:17.955 --> 00:01:23.859
They care about artists, copyright, security, creative control, and all the things you'd want them to care about.
16
00:01:24.179 --> 00:01:28.542
I actually think people would be impressed if they heard the conversations that were happening behind closed doors.
17
00:01:28.562 --> 00:01:28.662
Yeah.
18
00:01:29.062 --> 00:01:32.746
But the studios and the production companies, many of them just don't want to have the argument.
19
00:01:33.066 --> 00:01:33.687
And I get it.
20
00:01:33.847 --> 00:01:44.157
I get called a hypocrite on the internet most weeks, despite spending an unreasonable amount of time arguing that artists should be paid and AI companies shouldn't get to help themselves to everybody's work.
21
00:01:44.357 --> 00:01:48.461
So perhaps my faith in the general public's appetite for nuance is misplaced.
22
00:01:48.981 --> 00:01:53.106
Still, it creates this bizarre state where everybody knows something is happening.
23
00:01:53.446 --> 00:01:56.870
Lots of people are quietly doing it, but no one wants to talk about it.
24
00:01:57.310 --> 00:02:01.014
It's also why potential guests chicken out before coming on my podcast.
25
00:02:01.395 --> 00:02:07.341
I've lost count of the amount of times people have called me the day of and said, sorry, my company just won't let me talk to you.
26
00:02:07.982 --> 00:02:11.206
I actually find it a little bit flattering, like little old me.
27
00:02:11.386 --> 00:02:13.289
Which is why this week caught my attention.
28
00:02:13.629 --> 00:02:17.795
The public conversation finally starts to be catching up with the private experimentation.
29
00:02:18.255 --> 00:02:22.300
And some very accomplished creative people are starting to put their names on the poster.
30
00:02:23.001 --> 00:02:30.344
It seems that old man Hollywood is finally admitting that the young woman he keeps introducing as his assistant may in fact be his girlfriend.
31
00:02:30.764 --> 00:02:32.044
And at least that's honest.
32
00:02:32.564 --> 00:02:36.446
We covered the private Kathleen Kennedy and AFI discussions a few weeks ago.
33
00:02:36.586 --> 00:02:38.727
Of course filmmakers were experimenting.
34
00:02:39.027 --> 00:02:47.810
Filmmakers are professionally incapable of seeing a new tool and not asking whether it might help them move another rung up the social and professional ladder of Hollywood.
35
00:02:48.370 --> 00:02:50.191
I know from personal experience.
36
00:02:50.751 --> 00:02:53.937
But at least now we can argue about the things that people are actually making.
37
00:02:54.378 --> 00:02:55.079
Show me the work.
38
00:02:55.179 --> 00:02:56.041
Tell me what you did.
39
00:02:56.161 --> 00:02:57.082
Tell me what it cost.
40
00:02:57.503 --> 00:03:01.210
And tell me whether it allowed something to exist otherwise that wouldn't have.
41
00:03:02.392 --> 00:03:05.073
And then most importantly, was the thing any good?
42
00:03:05.293 --> 00:03:07.153
That's having a useful conversation.
43
00:03:07.793 --> 00:03:09.794
And so let's get into who started kicking this off.
44
00:03:09.874 --> 00:03:16.115
And it's two alums from the second golden age of animation at Disney, Jeffrey Katzenberg and Rob Minkoff.
45
00:03:16.255 --> 00:03:21.337
Jeffrey Katzenberg told The Wrap this week that he's developing plans for an AI-focused animation venture.
46
00:03:21.537 --> 00:03:23.537
The article was really light on specifics.
47
00:03:23.617 --> 00:03:31.139
Katzenberg ran Disney's film operation during the aforementioned second golden age of animation, and then founded DreamWorks Animation.
48
00:03:31.619 --> 00:03:35.882
He also gave us Quibi, which in fairness wasn't wrong that people wanted short-form video.
49
00:03:35.923 --> 00:03:37.764
It just got almost everything else wrong.
50
00:03:38.364 --> 00:03:43.989
Still, if Jeffrey Katzenberg thinks there might be a new animation business built around AI, I'm interested.
51
00:03:44.209 --> 00:03:45.850
And I think investors will be too.
52
00:03:46.391 --> 00:03:51.115
His argument is that lower costs could mean more stories and more projects actually getting financed.
53
00:03:51.635 --> 00:03:57.279
I've said on this show, the labor conversation often assumes a fixed amount of creative output.
54
00:03:57.799 --> 00:03:59.941
One movie used to require 500 people.
55
00:04:00.301 --> 00:04:03.583
Now it requires 300, therefore 200 people lose.
56
00:04:04.223 --> 00:04:06.505
That is, sure, one possible outcome.
57
00:04:06.805 --> 00:04:11.028
But what if reducing cost means five films get financed instead of one?
58
00:04:11.568 --> 00:04:16.732
What if stories that die in development because animation economics don't work suddenly become possible?
59
00:04:16.752 --> 00:04:17.112
I've said,
60
00:04:17.272 --> 00:04:20.497
I don't know which effect wins, efficiency or expansion.
61
00:04:20.877 --> 00:04:21.999
Neither does Katzenberg.
62
00:04:22.139 --> 00:04:23.801
But that's the experiment worth running.
63
00:04:24.061 --> 00:04:27.506
And then there's Rob Minkoff, who directed The Lion King under Katzenberg's Aegis.
64
00:04:28.367 --> 00:04:31.151
And he's now developing Storm Dogs with Particle 6.
65
00:04:31.411 --> 00:04:33.154
Rob was on my podcast last year.
66
00:04:33.254 --> 00:04:34.915
He's an early adopter of AI.
67
00:04:35.095 --> 00:04:46.463
And he points to the fact that his proposed production will use human performances for timing and expression, physicality and emotional choices, with technology translating those performances into finished characters.
68
00:04:46.704 --> 00:04:52.688
A synthetic character driven by an actor is not the same proposition as a synthetic character replacing the actor.
69
00:04:53.128 --> 00:05:00.734
It still raises questions around rights, compensation, and labor, particularly because Particle 6 is also the company behind Tilly Norwood.
70
00:05:01.014 --> 00:05:07.379
But at least we have an actual production to interrogate, and what they're proposing seems well within the SAG guidelines.
71
00:05:07.819 --> 00:05:13.904
Katzenberg and Minkoff know quite a lot about making audiences care about animated films and television.
72
00:05:14.344 --> 00:05:16.445
They have the receipts, so let's see what they can do.
73
00:05:16.465 --> 00:05:16.566
Okay.
74
00:05:17.226 --> 00:05:22.971
And this goes back to a story we covered before, which is Ben Affleck's company Interpositive being bought by Netflix.
75
00:05:23.151 --> 00:05:28.956
And I know people at Interpositive, and the secrecy around who they were working for was properly hysterical.
76
00:05:29.496 --> 00:05:34.160
For the first few conversations, they wouldn't even confirm who their Oscar-winning backer was.
77
00:05:34.540 --> 00:05:37.883
It was government-level secrecy until the second drink hit.
78
00:05:38.323 --> 00:05:43.089
But, you know, Netflix bought the company for roughly $587 million in cash.
79
00:05:43.469 --> 00:05:47.874
So while the public conversation was still stuck on, should filmmakers even use AI?
80
00:05:48.235 --> 00:05:54.322
Somewhere else in Hollywood, in a conference room high up in Netflix Tower, the actual question was, how's half a billion?
81
00:05:54.522 --> 00:05:58.606
Darren Aronofsky was also one of the first men through the door, and he got punched in the face for it.
82
00:05:59.087 --> 00:06:09.869
You know, he created Primordial Suit and then publicly started making On This Day 1776 with Time Studios, where AI-generated imagery sits alongside human writers, actors, music, and VFX.
83
00:06:10.369 --> 00:06:11.189
And the reaction?
84
00:06:11.309 --> 00:06:12.970
Well, a lot of people called it slop.
85
00:06:13.590 --> 00:06:17.270
Slop is to AI what disruptive was to startups in 2024.
86
00:06:17.350 --> 00:06:19.871
It must legally appear in every conversation.
87
00:06:20.211 --> 00:06:22.731
But I respect Aronofsky for putting his name on the work.
88
00:06:23.051 --> 00:06:23.891
You don't have to like it.
89
00:06:23.951 --> 00:06:24.752
I may not like it.
90
00:06:24.792 --> 00:06:26.332
But that's kind of beside the point.
91
00:06:26.812 --> 00:06:30.413
He's publicly testing a proposition and allowing people to judge the result.
92
00:06:30.994 --> 00:06:32.994
That's the creative industries as they've always been.
93
00:06:33.655 --> 00:06:34.495
Let's make the thing.
94
00:06:34.775 --> 00:06:35.695
Let's look at the thing.
95
00:06:36.016 --> 00:06:38.236
Then we can decide whether you think the thing is shit.
96
00:06:38.356 --> 00:06:41.898
And then on another production, 10 shooting days became two shooting days.
97
00:06:42.538 --> 00:06:53.806
At the Creative Investors Conference in San Sebastian, Goodfellas CEO Vincent Maraval discussed AI use on, and I'm gonna screw up this name, Michael Hazinovickius' The Third Hand.
98
00:06:54.247 --> 00:07:00.772
According to Maraval, a major battle sequence that would traditionally have required about 10 shooting days was completed in two.
99
00:07:00.872 --> 00:07:08.297
Before everybody turns to the film is now 80% cheaper line, this is one sequence on one film, so everybody calm down.
100
00:07:08.677 --> 00:07:09.978
But it does give us an actual example.
101
00:07:09.998 --> 00:07:10.799
It's a very, very, very, very, very,
102
00:07:10.899 --> 00:07:19.410
If the movie was already financed and definitely happening, eight fewer shooting days potentially means eight fewer days of employment for a lot of people, so that's fair.
103
00:07:19.731 --> 00:07:25.959
If the cost of those 10 days would have compromised the sequence or stopped the film from being financed, the calculation changes.
104
00:07:25.979 --> 00:07:26.099
But...
105
00:07:26.620 --> 00:07:29.583
There's another thing people outside production often miss.
106
00:07:30.004 --> 00:07:33.668
Film and commercial budgets are largely exercises in allocation.
107
00:07:34.068 --> 00:07:37.492
The goal is not to finish with a giant pile of money left over.
108
00:07:37.893 --> 00:07:45.061
The goal is to deliver the thing on budget and achieve the most amount of value so that it can become profitable when you sell it.
109
00:07:45.281 --> 00:07:55.747
So if you save money on a battle sequence, you may spend it on another shooting day, a better location, more time with the cast, VFX, music, finishing, or a sequence that otherwise gets cut.
110
00:07:56.267 --> 00:08:01.850
So eight fewer shooting days doesn't automatically mean the value of eight shooting days simply vanishes.
111
00:08:02.391 --> 00:08:05.793
And Maravelle's broader argument is also the one that I made last week.
112
00:08:06.313 --> 00:08:10.575
Sometimes the relevant comparison isn't 300 jobs versus 100 jobs.
113
00:08:10.895 --> 00:08:13.117
It's 100 jobs versus no film at all.
114
00:08:13.937 --> 00:08:22.362
If technology brings a project inside the line where somebody will finance it, that's a production that exists, job that exists, and a film that gets made.
115
00:08:22.582 --> 00:08:24.644
The question's whether that is actually what happens.
116
00:08:24.744 --> 00:08:27.545
Does the saving create more ambition inside the same movie?
117
00:08:28.066 --> 00:08:31.348
Does it allow the film fund to greenlight another project?
118
00:08:31.728 --> 00:08:33.549
Does it just reduce the cost base?
119
00:08:33.749 --> 00:08:34.810
Those are the questions.
120
00:08:35.170 --> 00:08:36.491
Let's start talking about them.
121
00:08:36.891 --> 00:08:39.574
And that's why I like having an actual example on the table.
122
00:08:39.874 --> 00:08:42.538
We don't need to dismiss the efficiency argument.
123
00:08:42.938 --> 00:08:44.560
We don't need to swallow it whole.
124
00:08:44.940 --> 00:08:48.284
We can finally start looking at the economics and asking what happens next.
125
00:08:48.564 --> 00:08:54.611
Again, this is what's valuable about having people talk about this out loud and not just hide it.
126
00:08:55.412 --> 00:09:01.734
And then publishing is even getting in the mix on it, saying, listen, we get it, you're going to use it, but please raise your hand and tell an adult when you do.
127
00:09:02.494 --> 00:09:07.815
Penguin Random House now has guidance for authors and illustrators who choose to use AI.
128
00:09:07.996 --> 00:09:11.436
That doesn't mean the PRH has stopped objecting to unauthorized training.
129
00:09:11.817 --> 00:09:13.937
It means the conversation is getting granular.
130
00:09:14.397 --> 00:09:15.197
What did you use?
131
00:09:15.458 --> 00:09:16.478
What did it contribute?
132
00:09:16.678 --> 00:09:17.818
What needs to be disclosed?
133
00:09:17.938 --> 00:09:19.239
And who owns the result?
134
00:09:19.839 --> 00:09:23.540
And then the music industry is getting in on the same thing, and they just want to get paid.
135
00:09:24.420 --> 00:09:29.706
For a while, there was something close to a don't-ask-don't-tell culture around AI in the music business.
136
00:09:30.066 --> 00:09:38.114
Producers, writers, and artists were experimenting privately while being understandably reluctant to announce it publicly and spend three days being screamed at online.
137
00:09:38.295 --> 00:09:44.902
Universal Music Group and Eleven Labs are building a licensed AI music creation platform involving participating artists and songwriters.
138
00:09:45.322 --> 00:09:46.143
That word matters.
139
00:09:46.263 --> 00:09:46.864
Licensed.
140
00:09:47.364 --> 00:09:56.448
A few weeks ago, we talked about Sony and Warner simultaneously investing in licensed generative AI and also suing Anthropic over alleged unauthorized use of music.
141
00:09:56.848 --> 00:09:58.689
People acted as though that was contradictory.
142
00:09:58.969 --> 00:09:59.429
It wasn't.
143
00:09:59.889 --> 00:10:03.371
The emerging creative industry position may turn out to be incredibly simple.
144
00:10:03.811 --> 00:10:07.274
You can build the thing, you just don't automatically get our shit for free.
145
00:10:07.695 --> 00:10:14.902
Universal appears interested in what happens when fans can remix, reinterpret, and create with artists' works under an actual commercial framework.
146
00:10:15.302 --> 00:10:16.163
I can see the appeal.
147
00:10:16.483 --> 00:10:18.305
Music has always invited participation.
148
00:10:18.345 --> 00:10:21.628
We sing it, cover it, remix it, and murder it at karaoke.
149
00:10:22.149 --> 00:10:25.412
So why wouldn't new technology create new forms of participation?
150
00:10:25.532 --> 00:10:26.012
Sounds fun!
151
00:10:26.693 --> 00:10:30.976
But I also want to see the contract, because license tells me somebody is being paid.
152
00:10:31.316 --> 00:10:35.579
It doesn't yet tell me who, how much, or how much control the artist actually has.
153
00:10:35.919 --> 00:10:36.660
Can they say no?
154
00:10:37.100 --> 00:10:38.541
Can they restrict certain uses?
155
00:10:38.861 --> 00:10:40.382
How does the songwriter participate?
156
00:10:40.462 --> 00:10:42.323
Does the artist get meaningful economics?
157
00:10:42.464 --> 00:10:46.146
Those questions aren't anti-innovation, they're the business model of creativity.
158
00:10:46.286 --> 00:10:48.227
Advertising may actually be furthest along.
159
00:10:48.367 --> 00:10:49.548
Agencies are using AI.
160
00:10:49.568 --> 00:10:50.849
The surprise there is gone.
161
00:10:51.309 --> 00:10:54.652
The new complaint, though, is that work is sort of becoming good enough...
162
00:10:55.232 --> 00:10:56.694
I mean, not great.
163
00:10:57.155 --> 00:11:01.441
Digiday reported this week on Marketeers complaining about a good enough and go culture.
164
00:11:01.862 --> 00:11:04.986
And I think it's one of the most important creative consequences in all of this.
165
00:11:05.126 --> 00:11:10.494
We've created astonishing technology and just found a way to send the client the wrong logo faster and more efficiently.
166
00:11:11.215 --> 00:11:12.897
20 executions became 200.
167
00:11:13.178 --> 00:11:15.420
The productivity dashboard looks amazing.
168
00:11:15.740 --> 00:11:19.164
And then somebody asks, did anybody remember this piece of shit?
169
00:11:19.865 --> 00:11:27.094
We're becoming extraordinarily efficient at producing things people can ignore, which is why I remain bullish on truly great creative people.
170
00:11:27.114 --> 00:11:27.194
And,
171
00:11:27.594 --> 00:11:31.395
If competent execution becomes abundant, competent execution becomes cheap.
172
00:11:31.856 --> 00:11:33.436
The scarce thing moves upstream.
173
00:11:33.796 --> 00:11:40.358
Taste and judgment and concept and cultural instinct, storytelling, and knowing crucially what not to make.
174
00:11:40.559 --> 00:11:44.140
AI will make mediocrity incredibly scalable, just like YouTube did.
175
00:11:44.560 --> 00:11:46.601
But the question is, can it make me give a shit?
176
00:11:46.961 --> 00:11:49.342
And that's the bar and it's the only one that matters.
177
00:11:49.902 --> 00:11:52.843
But it creates another problem the industry needs to think seriously about.
178
00:11:53.403 --> 00:11:59.889
Great senior creative judgment doesn't just emerge from the womb wearing a black t-shirt and complaining about kerning.
179
00:12:00.429 --> 00:12:06.594
People get good through reps, junior work, mistakes, watching people who are better than them make things.
180
00:12:07.175 --> 00:12:20.546
If we automate away all the work through which the next generation learns, then congratulate ourselves because senior creative people remain so valuable, we may eventually discover that we forgot to manufacture any more senior creative people.
181
00:12:21.307 --> 00:12:26.108
An industry can save money today and accidentally stop training the people it'll need tomorrow.
182
00:12:26.188 --> 00:12:32.170
So, I don't know, maybe let the intern make a few things once in a while, or better yet, get more interns.
183
00:12:32.570 --> 00:12:35.371
And then YouTube would like you to have a word with your own timeline.
184
00:12:35.911 --> 00:12:41.273
YouTube announced a Gemini-powered conversational editing assistant for Shorts and YouTube Create.
185
00:12:41.413 --> 00:12:48.475
Creators will be able to upload footage, ask for an initial edit in ordinary language, and then move back to a normal timeline and change it manually.
186
00:12:48.575 --> 00:12:51.757
This is one of those AI products where the usefulness is immediately obvious.
187
00:12:51.937 --> 00:12:58.100
I used to be an editor and I can tell you, getting footage into a workable first assembly can consume a ridiculous amount of time.
188
00:12:58.761 --> 00:13:06.465
If software can find takes, cut pauses, change pacing, add text, sync music, and get me to a version I can react to, I kinda wanna try it.
189
00:13:07.045 --> 00:13:09.647
The important bit is that it hands you back the timeline.
190
00:13:10.067 --> 00:13:12.970
I don't want an AI editor that decides what my film is.
191
00:13:13.310 --> 00:13:16.753
I want something that gets me to the point where I can start making decisions faster.
192
00:13:17.113 --> 00:13:19.315
That distinction between assistants and authorship.
193
00:13:19.515 --> 00:13:24.140
When I was an editor's assistant, I did that for the editors so that they can go on and do their job.
194
00:13:24.580 --> 00:13:29.104
Now, every editor, no matter how senior you are, gets their own digital assistant.
195
00:13:29.304 --> 00:13:31.125
And that's not necessarily a bad thing.
196
00:13:31.185 --> 00:13:34.948
And then another open AI agent crossed a boundary, and apparently it wasn't alone.
197
00:13:35.428 --> 00:13:45.315
Australia revealed this week that an open AI agent gained unauthorized access to a Medicare statistics portal in June while carrying out what authorities describe as a benign research task.
198
00:13:45.555 --> 00:13:53.338
It accessed public and non-private material, although the government says individual Australians' personal medical records were not exposed.
199
00:13:53.798 --> 00:13:58.179
OpenAI discovered the incident later, and Australia says it was not notified until September.
200
00:13:58.279 --> 00:14:09.463
And this story is growing because OpenAI now says that dozens of third parties globally have been affected by autonomous agents bypassing security controls or otherwise negatively impacting their systems.
201
00:14:09.643 --> 00:14:14.585
The LA Times reported just today that that number may not be dozens of third parties globally.
202
00:14:14.866 --> 00:14:16.466
It may be tens of thousands.
203
00:14:16.847 --> 00:14:18.427
So this is a developing story.
204
00:14:18.668 --> 00:14:28.432
Australian reporting also found evidence of agents spending days trying different approaches to obtain other government health data, although those incidents have not been formally linked to the Medicare breach.
205
00:14:28.613 --> 00:14:31.254
That makes this particularly uncomfortable after last week.
206
00:14:31.614 --> 00:14:37.135
We had Hugging Faced, we had Gemini wandering into the public internet during a controlled cybersecurity exercise.
207
00:14:37.555 --> 00:14:43.216
Now we have another agent, this time carrying out a completely ordinary research task, crossing a real government boundary.
208
00:14:43.376 --> 00:14:46.257
And then, surprise, your chatbot might have been another chatbot.
209
00:14:46.457 --> 00:14:49.718
Anthropic's September threat report contains an extraordinary allegation.
210
00:14:50.218 --> 00:15:00.080
It says that Alibaba generated more than 151 million Claude exchanges between May and July as part of a large-scale effort to distill Claude's capabilities into its own models.
211
00:15:00.820 --> 00:15:12.570
It also alleges that Moonshot and DeepSeek silently routed some customer requests through Claude, showed Claude's answers to users as though they came from their own systems, and retained some exchanges for training.
212
00:15:13.111 --> 00:15:19.856
Anthropix says that some of the relayed material included internal company information, live credentials, and other sensitive data.
213
00:15:20.197 --> 00:15:23.700
These are Anthropix allegations and attributions, not adjudicated findings.
214
00:15:24.220 --> 00:15:25.880
The obvious story is competitive.
215
00:15:26.040 --> 00:15:28.101
AI company spends billions making model.
216
00:15:28.721 --> 00:15:31.522
Rival would quite like that model's homework to copy off of.
217
00:15:31.982 --> 00:15:36.423
But the confidentiality problem may be more important for anyone using these systems professionally.
218
00:15:36.883 --> 00:15:47.366
If I choose a provider because I believe I understand where my information is going, then discover my prompt was quietly forwarded to another model entirely, that arrangement has changed rather dramatically.
219
00:15:48.026 --> 00:15:58.374
Imagine putting an unreleased campaign, script, contract, source code, or company strategy into one provider and discovering it had taken an unexpected international holiday.
220
00:15:58.514 --> 00:16:05.820
And this is another reason that your company's enterprise AI strategy just can't be, just pick whatever chatbot you like.
221
00:16:06.340 --> 00:16:08.001
Where your data goes matters.
222
00:16:08.301 --> 00:16:12.504
And then, and then finally, Ellison Gate might actually be over.
223
00:16:12.984 --> 00:16:17.587
California and 11 other states reached a settlement with Paramount over its Warner Bros.
224
00:16:17.607 --> 00:16:18.688
Discovery acquisition.
225
00:16:19.148 --> 00:16:30.255
The agreement includes a five-year commitment for the combined company to release 30 films a year for the first two years and 32 a year for the following three, with at least four independent films a year.
226
00:16:30.735 --> 00:16:39.403
Paramount has also committed to at least $1.5 billion in additional U.S. film production spending over its 2025 level during those five years.
227
00:16:39.683 --> 00:16:41.785
And these rulings have teeth, you know.
228
00:16:41.845 --> 00:16:54.757
If Paramount misses the annual film output requirement, the settlement calls for a Miramax divestiture and payments of $30 million per missed film into funds including union and health and retirement plans.
229
00:16:55.398 --> 00:16:59.682
Remember when Tom Cruise said that a combined Paramount-Warner could easily make 30 films a year?
230
00:17:00.122 --> 00:17:06.908
Remember me spending several thousand words arguing that perhaps Hollywood's greatest innovation would be making some more fucking movies?
231
00:17:06.968 --> 00:17:11.031
Well, film output is now effectively part of the regulatory settlement.
232
00:17:11.471 --> 00:17:13.613
It also connects back to the first half of this piece.
233
00:17:14.114 --> 00:17:21.460
If AI and other production technologies really do reduce the cost of making ambitious work, this is where the expansion argument gets tested—
234
00:17:21.920 --> 00:17:26.082
not another demo, in how many movies actually get commissioned and made.
235
00:17:26.682 --> 00:17:32.904
Apparently the radical new principle is, if you're gonna own two giant movie studios, please make some fucking movies.
236
00:17:33.624 --> 00:17:35.025
And I support this innovation.
237
00:17:35.305 --> 00:17:36.685
So what actually changed?
238
00:17:36.746 --> 00:17:42.748
The important thing this week is not that the creative industries have suddenly decided that AI is good, they certainly haven't.
239
00:17:43.308 --> 00:17:48.470
There are still real questions around copyright, consent, compensation, and all the other things that we bang on about here.
240
00:17:49.010 --> 00:17:52.354
Those questions aren't going away, and they shouldn't, and we're still going to talk about them.
241
00:17:52.775 --> 00:17:59.323
What has changed, however, is that more creative people are finally willing to say publicly, yes, I use these tools.
242
00:17:59.964 --> 00:18:07.633
And that matters because it's almost impossible to have any serious conversation about creative quality when nobody will admit what they're actually doing.
243
00:18:08.274 --> 00:18:10.936
You can argue about ethics in the abstract forever.
244
00:18:10.996 --> 00:18:14.498
You can argue about policy and training data and labor impact.
245
00:18:14.838 --> 00:18:18.580
But creative industries eventually come back to the same brutally simple question.
246
00:18:19.341 --> 00:18:20.341
Is the work any good?
247
00:18:21.022 --> 00:18:22.583
Did it move you or surprise you?
248
00:18:22.643 --> 00:18:24.624
Does it feel authored?
249
00:18:24.904 --> 00:18:26.365
Is it worth your time?
250
00:18:26.505 --> 00:18:31.328
We haven't arrived at the mature version of this conversation yet because everybody's too afraid of being canceled.
251
00:18:32.008 --> 00:18:46.975
But it does feel like a useful first step because canceling people every time they admit to using a particular filmmaking tool just isn't helping us figure out where this technology belongs, where it doesn't, and what good work actually looks like.
252
00:18:47.736 --> 00:18:52.098
If people are finally willing to step into the light and say, yes, I used AI.
253
00:18:52.478 --> 00:18:53.758
Here's what I used it for.
254
00:18:54.159 --> 00:18:55.119
Here's what I made.
255
00:18:55.539 --> 00:18:57.120
Here was the economic impact.
256
00:18:57.420 --> 00:18:58.681
Here was the human impact.
257
00:18:58.981 --> 00:19:00.502
Here was an unintended benefit.
258
00:19:01.182 --> 00:19:02.183
then good.
259
00:19:02.703 --> 00:19:07.527
Because some of my favorite days in creative companies were spent arguing about what was good and what wasn't.
260
00:19:08.147 --> 00:19:09.408
Those were the good old days.
261
00:19:09.748 --> 00:19:13.271
And we can have them back, just as long as the conversation keeps evolving.
262
00:19:13.871 --> 00:19:15.873
So thank you for listening this week.
263
00:19:16.233 --> 00:19:17.174
I really appreciate it.
264
00:19:17.194 --> 00:19:24.399
We'll be back with our interview episode tomorrow with Ryan Black and Kite Jene from Luma Labs.
265
00:19:24.519 --> 00:19:28.742
Kite is a Ford deployed creative, which is the first time I've heard any company use that.
266
00:19:29.202 --> 00:19:42.845
But it's, you know, what Luma does when people are coming on and using its models, it teaches them how to use the nose base and agent systems by having creatives that will actually deploy into your company.
267
00:19:43.525 --> 00:19:44.585
So it's a really unique model.
268
00:19:44.645 --> 00:19:46.446
It's sort of stolen from the Palantir model.
269
00:19:47.786 --> 00:19:49.366
And, yeah, they're doing amazing things.
270
00:19:49.426 --> 00:19:52.247
Luma has one of the best models on the market.
271
00:19:52.887 --> 00:19:53.847
I think you'll really enjoy it.
272
00:19:53.867 --> 00:19:54.947
It's a really good conversation.
273
00:19:55.327 --> 00:19:57.628
Thank you for listening and tune in next week.
1
00:00:00.051 --> 00:00:02.593
Hi, this is the Week in Review from Technically Creative.
2
00:00:02.793 --> 00:00:13.001
I'm Orlando Wood, and over the next 20 minutes, I'm going to give you an overview of what I felt the most important news was last week in the ever-increasingly interdependent worlds of tech and the creative industries.
3
00:00:13.121 --> 00:00:14.843
This is the Week in Review, September 28, 2026.
4
00:00:16.757 --> 00:00:24.103
This week, it seemed like creative leaders started coming out of their dark AI dens and declaring proudly, yeah, we're using AI.
5
00:00:24.183 --> 00:00:24.703
What of it?
6
00:00:24.903 --> 00:00:33.029
It reminds me that an exec at Warner Brothers once said to me, everybody says they hate AI, but everybody's coming back from the bathroom with really great ideas.
7
00:00:33.830 --> 00:00:39.413
And that may be the best description I've heard of the creative industry's relationship with AI over the last few years.
8
00:00:39.793 --> 00:00:41.194
Publicly, horror.
9
00:00:41.554 --> 00:00:44.255
Privately, lots of experimentation and interest.
10
00:00:44.735 --> 00:01:01.404
As someone who helps big companies adopt software that uses AI, for me it's felt a bit like getting to date the prettiest girl in high school, but she refuses to acknowledge you in the hallway, she tells you you can't tell your friends, but she will make out with you under the bleachers on Tuesdays and Wednesdays during study hall.
11
00:01:01.984 --> 00:01:03.785
A24 was one of our first clients.
12
00:01:04.106 --> 00:01:06.187
We're allowed to say we work with them colloquially.
13
00:01:06.427 --> 00:01:08.989
We're just not allowed to say what we actually do for them.
14
00:01:09.529 --> 00:01:17.775
And it's the same with most of our industry clients, which is strange because the companies we work with are incredibly thoughtful about how they use this technology.
15
00:01:17.955 --> 00:01:23.859
They care about artists, copyright, security, creative control, and all the things you'd want them to care about.
16
00:01:24.179 --> 00:01:28.542
I actually think people would be impressed if they heard the conversations that were happening behind closed doors.
17
00:01:28.562 --> 00:01:28.662
Yeah.
18
00:01:29.062 --> 00:01:32.746
But the studios and the production companies, many of them just don't want to have the argument.
19
00:01:33.066 --> 00:01:33.687
And I get it.
20
00:01:33.847 --> 00:01:44.157
I get called a hypocrite on the internet most weeks, despite spending an unreasonable amount of time arguing that artists should be paid and AI companies shouldn't get to help themselves to everybody's work.
21
00:01:44.357 --> 00:01:48.461
So perhaps my faith in the general public's appetite for nuance is misplaced.
22
00:01:48.981 --> 00:01:53.106
Still, it creates this bizarre state where everybody knows something is happening.
23
00:01:53.446 --> 00:01:56.870
Lots of people are quietly doing it, but no one wants to talk about it.
24
00:01:57.310 --> 00:02:01.014
It's also why potential guests chicken out before coming on my podcast.
25
00:02:01.395 --> 00:02:07.341
I've lost count of the amount of times people have called me the day of and said, sorry, my company just won't let me talk to you.
26
00:02:07.982 --> 00:02:11.206
I actually find it a little bit flattering, like little old me.
27
00:02:11.386 --> 00:02:13.289
Which is why this week caught my attention.
28
00:02:13.629 --> 00:02:17.795
The public conversation finally starts to be catching up with the private experimentation.
29
00:02:18.255 --> 00:02:22.300
And some very accomplished creative people are starting to put their names on the poster.
30
00:02:23.001 --> 00:02:30.344
It seems that old man Hollywood is finally admitting that the young woman he keeps introducing as his assistant may in fact be his girlfriend.
31
00:02:30.764 --> 00:02:32.044
And at least that's honest.
32
00:02:32.564 --> 00:02:36.446
We covered the private Kathleen Kennedy and AFI discussions a few weeks ago.
33
00:02:36.586 --> 00:02:38.727
Of course filmmakers were experimenting.
34
00:02:39.027 --> 00:02:47.810
Filmmakers are professionally incapable of seeing a new tool and not asking whether it might help them move another rung up the social and professional ladder of Hollywood.
35
00:02:48.370 --> 00:02:50.191
I know from personal experience.
36
00:02:50.751 --> 00:02:53.937
But at least now we can argue about the things that people are actually making.
37
00:02:54.378 --> 00:02:55.079
Show me the work.
38
00:02:55.179 --> 00:02:56.041
Tell me what you did.
39
00:02:56.161 --> 00:02:57.082
Tell me what it cost.
40
00:02:57.503 --> 00:03:01.210
And tell me whether it allowed something to exist otherwise that wouldn't have.
41
00:03:02.392 --> 00:03:05.073
And then most importantly, was the thing any good?
42
00:03:05.293 --> 00:03:07.153
That's having a useful conversation.
43
00:03:07.793 --> 00:03:09.794
And so let's get into who started kicking this off.
44
00:03:09.874 --> 00:03:16.115
And it's two alums from the second golden age of animation at Disney, Jeffrey Katzenberg and Rob Minkoff.
45
00:03:16.255 --> 00:03:21.337
Jeffrey Katzenberg told The Wrap this week that he's developing plans for an AI-focused animation venture.
46
00:03:21.537 --> 00:03:23.537
The article was really light on specifics.
47
00:03:23.617 --> 00:03:31.139
Katzenberg ran Disney's film operation during the aforementioned second golden age of animation, and then founded DreamWorks Animation.
48
00:03:31.619 --> 00:03:35.882
He also gave us Quibi, which in fairness wasn't wrong that people wanted short-form video.
49
00:03:35.923 --> 00:03:37.764
It just got almost everything else wrong.
50
00:03:38.364 --> 00:03:43.989
Still, if Jeffrey Katzenberg thinks there might be a new animation business built around AI, I'm interested.
51
00:03:44.209 --> 00:03:45.850
And I think investors will be too.
52
00:03:46.391 --> 00:03:51.115
His argument is that lower costs could mean more stories and more projects actually getting financed.
53
00:03:51.635 --> 00:03:57.279
I've said on this show, the labor conversation often assumes a fixed amount of creative output.
54
00:03:57.799 --> 00:03:59.941
One movie used to require 500 people.
55
00:04:00.301 --> 00:04:03.583
Now it requires 300, therefore 200 people lose.
56
00:04:04.223 --> 00:04:06.505
That is, sure, one possible outcome.
57
00:04:06.805 --> 00:04:11.028
But what if reducing cost means five films get financed instead of one?
58
00:04:11.568 --> 00:04:16.732
What if stories that die in development because animation economics don't work suddenly become possible?
59
00:04:16.752 --> 00:04:17.112
I've said,
60
00:04:17.272 --> 00:04:20.497
I don't know which effect wins, efficiency or expansion.
61
00:04:20.877 --> 00:04:21.999
Neither does Katzenberg.
62
00:04:22.139 --> 00:04:23.801
But that's the experiment worth running.
63
00:04:24.061 --> 00:04:27.506
And then there's Rob Minkoff, who directed The Lion King under Katzenberg's Aegis.
64
00:04:28.367 --> 00:04:31.151
And he's now developing Storm Dogs with Particle 6.
65
00:04:31.411 --> 00:04:33.154
Rob was on my podcast last year.
66
00:04:33.254 --> 00:04:34.915
He's an early adopter of AI.
67
00:04:35.095 --> 00:04:46.463
And he points to the fact that his proposed production will use human performances for timing and expression, physicality and emotional choices, with technology translating those performances into finished characters.
68
00:04:46.704 --> 00:04:52.688
A synthetic character driven by an actor is not the same proposition as a synthetic character replacing the actor.
69
00:04:53.128 --> 00:05:00.734
It still raises questions around rights, compensation, and labor, particularly because Particle 6 is also the company behind Tilly Norwood.
70
00:05:01.014 --> 00:05:07.379
But at least we have an actual production to interrogate, and what they're proposing seems well within the SAG guidelines.
71
00:05:07.819 --> 00:05:13.904
Katzenberg and Minkoff know quite a lot about making audiences care about animated films and television.
72
00:05:14.344 --> 00:05:16.445
They have the receipts, so let's see what they can do.
73
00:05:16.465 --> 00:05:16.566
Okay.
74
00:05:17.226 --> 00:05:22.971
And this goes back to a story we covered before, which is Ben Affleck's company Interpositive being bought by Netflix.
75
00:05:23.151 --> 00:05:28.956
And I know people at Interpositive, and the secrecy around who they were working for was properly hysterical.
76
00:05:29.496 --> 00:05:34.160
For the first few conversations, they wouldn't even confirm who their Oscar-winning backer was.
77
00:05:34.540 --> 00:05:37.883
It was government-level secrecy until the second drink hit.
78
00:05:38.323 --> 00:05:43.089
But, you know, Netflix bought the company for roughly $587 million in cash.
79
00:05:43.469 --> 00:05:47.874
So while the public conversation was still stuck on, should filmmakers even use AI?
80
00:05:48.235 --> 00:05:54.322
Somewhere else in Hollywood, in a conference room high up in Netflix Tower, the actual question was, how's half a billion?
81
00:05:54.522 --> 00:05:58.606
Darren Aronofsky was also one of the first men through the door, and he got punched in the face for it.
82
00:05:59.087 --> 00:06:09.869
You know, he created Primordial Suit and then publicly started making On This Day 1776 with Time Studios, where AI-generated imagery sits alongside human writers, actors, music, and VFX.
83
00:06:10.369 --> 00:06:11.189
And the reaction?
84
00:06:11.309 --> 00:06:12.970
Well, a lot of people called it slop.
85
00:06:13.590 --> 00:06:17.270
Slop is to AI what disruptive was to startups in 2024.
86
00:06:17.350 --> 00:06:19.871
It must legally appear in every conversation.
87
00:06:20.211 --> 00:06:22.731
But I respect Aronofsky for putting his name on the work.
88
00:06:23.051 --> 00:06:23.891
You don't have to like it.
89
00:06:23.951 --> 00:06:24.752
I may not like it.
90
00:06:24.792 --> 00:06:26.332
But that's kind of beside the point.
91
00:06:26.812 --> 00:06:30.413
He's publicly testing a proposition and allowing people to judge the result.
92
00:06:30.994 --> 00:06:32.994
That's the creative industries as they've always been.
93
00:06:33.655 --> 00:06:34.495
Let's make the thing.
94
00:06:34.775 --> 00:06:35.695
Let's look at the thing.
95
00:06:36.016 --> 00:06:38.236
Then we can decide whether you think the thing is shit.
96
00:06:38.356 --> 00:06:41.898
And then on another production, 10 shooting days became two shooting days.
97
00:06:42.538 --> 00:06:53.806
At the Creative Investors Conference in San Sebastian, Goodfellas CEO Vincent Maraval discussed AI use on, and I'm gonna screw up this name, Michael Hazinovickius' The Third Hand.
98
00:06:54.247 --> 00:07:00.772
According to Maraval, a major battle sequence that would traditionally have required about 10 shooting days was completed in two.
99
00:07:00.872 --> 00:07:08.297
Before everybody turns to the film is now 80% cheaper line, this is one sequence on one film, so everybody calm down.
100
00:07:08.677 --> 00:07:09.978
But it does give us an actual example.
101
00:07:09.998 --> 00:07:10.799
It's a very, very, very, very, very,
102
00:07:10.899 --> 00:07:19.410
If the movie was already financed and definitely happening, eight fewer shooting days potentially means eight fewer days of employment for a lot of people, so that's fair.
103
00:07:19.731 --> 00:07:25.959
If the cost of those 10 days would have compromised the sequence or stopped the film from being financed, the calculation changes.
104
00:07:25.979 --> 00:07:26.099
But...
105
00:07:26.620 --> 00:07:29.583
There's another thing people outside production often miss.
106
00:07:30.004 --> 00:07:33.668
Film and commercial budgets are largely exercises in allocation.
107
00:07:34.068 --> 00:07:37.492
The goal is not to finish with a giant pile of money left over.
108
00:07:37.893 --> 00:07:45.061
The goal is to deliver the thing on budget and achieve the most amount of value so that it can become profitable when you sell it.
109
00:07:45.281 --> 00:07:55.747
So if you save money on a battle sequence, you may spend it on another shooting day, a better location, more time with the cast, VFX, music, finishing, or a sequence that otherwise gets cut.
110
00:07:56.267 --> 00:08:01.850
So eight fewer shooting days doesn't automatically mean the value of eight shooting days simply vanishes.
111
00:08:02.391 --> 00:08:05.793
And Maravelle's broader argument is also the one that I made last week.
112
00:08:06.313 --> 00:08:10.575
Sometimes the relevant comparison isn't 300 jobs versus 100 jobs.
113
00:08:10.895 --> 00:08:13.117
It's 100 jobs versus no film at all.
114
00:08:13.937 --> 00:08:22.362
If technology brings a project inside the line where somebody will finance it, that's a production that exists, job that exists, and a film that gets made.
115
00:08:22.582 --> 00:08:24.644
The question's whether that is actually what happens.
116
00:08:24.744 --> 00:08:27.545
Does the saving create more ambition inside the same movie?
117
00:08:28.066 --> 00:08:31.348
Does it allow the film fund to greenlight another project?
118
00:08:31.728 --> 00:08:33.549
Does it just reduce the cost base?
119
00:08:33.749 --> 00:08:34.810
Those are the questions.
120
00:08:35.170 --> 00:08:36.491
Let's start talking about them.
121
00:08:36.891 --> 00:08:39.574
And that's why I like having an actual example on the table.
122
00:08:39.874 --> 00:08:42.538
We don't need to dismiss the efficiency argument.
123
00:08:42.938 --> 00:08:44.560
We don't need to swallow it whole.
124
00:08:44.940 --> 00:08:48.284
We can finally start looking at the economics and asking what happens next.
125
00:08:48.564 --> 00:08:54.611
Again, this is what's valuable about having people talk about this out loud and not just hide it.
126
00:08:55.412 --> 00:09:01.734
And then publishing is even getting in the mix on it, saying, listen, we get it, you're going to use it, but please raise your hand and tell an adult when you do.
127
00:09:02.494 --> 00:09:07.815
Penguin Random House now has guidance for authors and illustrators who choose to use AI.
128
00:09:07.996 --> 00:09:11.436
That doesn't mean the PRH has stopped objecting to unauthorized training.
129
00:09:11.817 --> 00:09:13.937
It means the conversation is getting granular.
130
00:09:14.397 --> 00:09:15.197
What did you use?
131
00:09:15.458 --> 00:09:16.478
What did it contribute?
132
00:09:16.678 --> 00:09:17.818
What needs to be disclosed?
133
00:09:17.938 --> 00:09:19.239
And who owns the result?
134
00:09:19.839 --> 00:09:23.540
And then the music industry is getting in on the same thing, and they just want to get paid.
135
00:09:24.420 --> 00:09:29.706
For a while, there was something close to a don't-ask-don't-tell culture around AI in the music business.
136
00:09:30.066 --> 00:09:38.114
Producers, writers, and artists were experimenting privately while being understandably reluctant to announce it publicly and spend three days being screamed at online.
137
00:09:38.295 --> 00:09:44.902
Universal Music Group and Eleven Labs are building a licensed AI music creation platform involving participating artists and songwriters.
138
00:09:45.322 --> 00:09:46.143
That word matters.
139
00:09:46.263 --> 00:09:46.864
Licensed.
140
00:09:47.364 --> 00:09:56.448
A few weeks ago, we talked about Sony and Warner simultaneously investing in licensed generative AI and also suing Anthropic over alleged unauthorized use of music.
141
00:09:56.848 --> 00:09:58.689
People acted as though that was contradictory.
142
00:09:58.969 --> 00:09:59.429
It wasn't.
143
00:09:59.889 --> 00:10:03.371
The emerging creative industry position may turn out to be incredibly simple.
144
00:10:03.811 --> 00:10:07.274
You can build the thing, you just don't automatically get our shit for free.
145
00:10:07.695 --> 00:10:14.902
Universal appears interested in what happens when fans can remix, reinterpret, and create with artists' works under an actual commercial framework.
146
00:10:15.302 --> 00:10:16.163
I can see the appeal.
147
00:10:16.483 --> 00:10:18.305
Music has always invited participation.
148
00:10:18.345 --> 00:10:21.628
We sing it, cover it, remix it, and murder it at karaoke.
149
00:10:22.149 --> 00:10:25.412
So why wouldn't new technology create new forms of participation?
150
00:10:25.532 --> 00:10:26.012
Sounds fun!
151
00:10:26.693 --> 00:10:30.976
But I also want to see the contract, because license tells me somebody is being paid.
152
00:10:31.316 --> 00:10:35.579
It doesn't yet tell me who, how much, or how much control the artist actually has.
153
00:10:35.919 --> 00:10:36.660
Can they say no?
154
00:10:37.100 --> 00:10:38.541
Can they restrict certain uses?
155
00:10:38.861 --> 00:10:40.382
How does the songwriter participate?
156
00:10:40.462 --> 00:10:42.323
Does the artist get meaningful economics?
157
00:10:42.464 --> 00:10:46.146
Those questions aren't anti-innovation, they're the business model of creativity.
158
00:10:46.286 --> 00:10:48.227
Advertising may actually be furthest along.
159
00:10:48.367 --> 00:10:49.548
Agencies are using AI.
160
00:10:49.568 --> 00:10:50.849
The surprise there is gone.
161
00:10:51.309 --> 00:10:54.652
The new complaint, though, is that work is sort of becoming good enough...
162
00:10:55.232 --> 00:10:56.694
I mean, not great.
163
00:10:57.155 --> 00:11:01.441
Digiday reported this week on Marketeers complaining about a good enough and go culture.
164
00:11:01.862 --> 00:11:04.986
And I think it's one of the most important creative consequences in all of this.
165
00:11:05.126 --> 00:11:10.494
We've created astonishing technology and just found a way to send the client the wrong logo faster and more efficiently.
166
00:11:11.215 --> 00:11:12.897
20 executions became 200.
167
00:11:13.178 --> 00:11:15.420
The productivity dashboard looks amazing.
168
00:11:15.740 --> 00:11:19.164
And then somebody asks, did anybody remember this piece of shit?
169
00:11:19.865 --> 00:11:27.094
We're becoming extraordinarily efficient at producing things people can ignore, which is why I remain bullish on truly great creative people.
170
00:11:27.114 --> 00:11:27.194
And,
171
00:11:27.594 --> 00:11:31.395
If competent execution becomes abundant, competent execution becomes cheap.
172
00:11:31.856 --> 00:11:33.436
The scarce thing moves upstream.
173
00:11:33.796 --> 00:11:40.358
Taste and judgment and concept and cultural instinct, storytelling, and knowing crucially what not to make.
174
00:11:40.559 --> 00:11:44.140
AI will make mediocrity incredibly scalable, just like YouTube did.
175
00:11:44.560 --> 00:11:46.601
But the question is, can it make me give a shit?
176
00:11:46.961 --> 00:11:49.342
And that's the bar and it's the only one that matters.
177
00:11:49.902 --> 00:11:52.843
But it creates another problem the industry needs to think seriously about.
178
00:11:53.403 --> 00:11:59.889
Great senior creative judgment doesn't just emerge from the womb wearing a black t-shirt and complaining about kerning.
179
00:12:00.429 --> 00:12:06.594
People get good through reps, junior work, mistakes, watching people who are better than them make things.
180
00:12:07.175 --> 00:12:20.546
If we automate away all the work through which the next generation learns, then congratulate ourselves because senior creative people remain so valuable, we may eventually discover that we forgot to manufacture any more senior creative people.
181
00:12:21.307 --> 00:12:26.108
An industry can save money today and accidentally stop training the people it'll need tomorrow.
182
00:12:26.188 --> 00:12:32.170
So, I don't know, maybe let the intern make a few things once in a while, or better yet, get more interns.
183
00:12:32.570 --> 00:12:35.371
And then YouTube would like you to have a word with your own timeline.
184
00:12:35.911 --> 00:12:41.273
YouTube announced a Gemini-powered conversational editing assistant for Shorts and YouTube Create.
185
00:12:41.413 --> 00:12:48.475
Creators will be able to upload footage, ask for an initial edit in ordinary language, and then move back to a normal timeline and change it manually.
186
00:12:48.575 --> 00:12:51.757
This is one of those AI products where the usefulness is immediately obvious.
187
00:12:51.937 --> 00:12:58.100
I used to be an editor and I can tell you, getting footage into a workable first assembly can consume a ridiculous amount of time.
188
00:12:58.761 --> 00:13:06.465
If software can find takes, cut pauses, change pacing, add text, sync music, and get me to a version I can react to, I kinda wanna try it.
189
00:13:07.045 --> 00:13:09.647
The important bit is that it hands you back the timeline.
190
00:13:10.067 --> 00:13:12.970
I don't want an AI editor that decides what my film is.
191
00:13:13.310 --> 00:13:16.753
I want something that gets me to the point where I can start making decisions faster.
192
00:13:17.113 --> 00:13:19.315
That distinction between assistants and authorship.
193
00:13:19.515 --> 00:13:24.140
When I was an editor's assistant, I did that for the editors so that they can go on and do their job.
194
00:13:24.580 --> 00:13:29.104
Now, every editor, no matter how senior you are, gets their own digital assistant.
195
00:13:29.304 --> 00:13:31.125
And that's not necessarily a bad thing.
196
00:13:31.185 --> 00:13:34.948
And then another open AI agent crossed a boundary, and apparently it wasn't alone.
197
00:13:35.428 --> 00:13:45.315
Australia revealed this week that an open AI agent gained unauthorized access to a Medicare statistics portal in June while carrying out what authorities describe as a benign research task.
198
00:13:45.555 --> 00:13:53.338
It accessed public and non-private material, although the government says individual Australians' personal medical records were not exposed.
199
00:13:53.798 --> 00:13:58.179
OpenAI discovered the incident later, and Australia says it was not notified until September.
200
00:13:58.279 --> 00:14:09.463
And this story is growing because OpenAI now says that dozens of third parties globally have been affected by autonomous agents bypassing security controls or otherwise negatively impacting their systems.
201
00:14:09.643 --> 00:14:14.585
The LA Times reported just today that that number may not be dozens of third parties globally.
202
00:14:14.866 --> 00:14:16.466
It may be tens of thousands.
203
00:14:16.847 --> 00:14:18.427
So this is a developing story.
204
00:14:18.668 --> 00:14:28.432
Australian reporting also found evidence of agents spending days trying different approaches to obtain other government health data, although those incidents have not been formally linked to the Medicare breach.
205
00:14:28.613 --> 00:14:31.254
That makes this particularly uncomfortable after last week.
206
00:14:31.614 --> 00:14:37.135
We had Hugging Faced, we had Gemini wandering into the public internet during a controlled cybersecurity exercise.
207
00:14:37.555 --> 00:14:43.216
Now we have another agent, this time carrying out a completely ordinary research task, crossing a real government boundary.
208
00:14:43.376 --> 00:14:46.257
And then, surprise, your chatbot might have been another chatbot.
209
00:14:46.457 --> 00:14:49.718
Anthropic's September threat report contains an extraordinary allegation.
210
00:14:50.218 --> 00:15:00.080
It says that Alibaba generated more than 151 million Claude exchanges between May and July as part of a large-scale effort to distill Claude's capabilities into its own models.
211
00:15:00.820 --> 00:15:12.570
It also alleges that Moonshot and DeepSeek silently routed some customer requests through Claude, showed Claude's answers to users as though they came from their own systems, and retained some exchanges for training.
212
00:15:13.111 --> 00:15:19.856
Anthropix says that some of the relayed material included internal company information, live credentials, and other sensitive data.
213
00:15:20.197 --> 00:15:23.700
These are Anthropix allegations and attributions, not adjudicated findings.
214
00:15:24.220 --> 00:15:25.880
The obvious story is competitive.
215
00:15:26.040 --> 00:15:28.101
AI company spends billions making model.
216
00:15:28.721 --> 00:15:31.522
Rival would quite like that model's homework to copy off of.
217
00:15:31.982 --> 00:15:36.423
But the confidentiality problem may be more important for anyone using these systems professionally.
218
00:15:36.883 --> 00:15:47.366
If I choose a provider because I believe I understand where my information is going, then discover my prompt was quietly forwarded to another model entirely, that arrangement has changed rather dramatically.
219
00:15:48.026 --> 00:15:58.374
Imagine putting an unreleased campaign, script, contract, source code, or company strategy into one provider and discovering it had taken an unexpected international holiday.
220
00:15:58.514 --> 00:16:05.820
And this is another reason that your company's enterprise AI strategy just can't be, just pick whatever chatbot you like.
221
00:16:06.340 --> 00:16:08.001
Where your data goes matters.
222
00:16:08.301 --> 00:16:12.504
And then, and then finally, Ellison Gate might actually be over.
223
00:16:12.984 --> 00:16:17.587
California and 11 other states reached a settlement with Paramount over its Warner Bros.
224
00:16:17.607 --> 00:16:18.688
Discovery acquisition.
225
00:16:19.148 --> 00:16:30.255
The agreement includes a five-year commitment for the combined company to release 30 films a year for the first two years and 32 a year for the following three, with at least four independent films a year.
226
00:16:30.735 --> 00:16:39.403
Paramount has also committed to at least $1.5 billion in additional U.S. film production spending over its 2025 level during those five years.
227
00:16:39.683 --> 00:16:41.785
And these rulings have teeth, you know.
228
00:16:41.845 --> 00:16:54.757
If Paramount misses the annual film output requirement, the settlement calls for a Miramax divestiture and payments of $30 million per missed film into funds including union and health and retirement plans.
229
00:16:55.398 --> 00:16:59.682
Remember when Tom Cruise said that a combined Paramount-Warner could easily make 30 films a year?
230
00:17:00.122 --> 00:17:06.908
Remember me spending several thousand words arguing that perhaps Hollywood's greatest innovation would be making some more fucking movies?
231
00:17:06.968 --> 00:17:11.031
Well, film output is now effectively part of the regulatory settlement.
232
00:17:11.471 --> 00:17:13.613
It also connects back to the first half of this piece.
233
00:17:14.114 --> 00:17:21.460
If AI and other production technologies really do reduce the cost of making ambitious work, this is where the expansion argument gets tested—
234
00:17:21.920 --> 00:17:26.082
not another demo, in how many movies actually get commissioned and made.
235
00:17:26.682 --> 00:17:32.904
Apparently the radical new principle is, if you're gonna own two giant movie studios, please make some fucking movies.
236
00:17:33.624 --> 00:17:35.025
And I support this innovation.
237
00:17:35.305 --> 00:17:36.685
So what actually changed?
238
00:17:36.746 --> 00:17:42.748
The important thing this week is not that the creative industries have suddenly decided that AI is good, they certainly haven't.
239
00:17:43.308 --> 00:17:48.470
There are still real questions around copyright, consent, compensation, and all the other things that we bang on about here.
240
00:17:49.010 --> 00:17:52.354
Those questions aren't going away, and they shouldn't, and we're still going to talk about them.
241
00:17:52.775 --> 00:17:59.323
What has changed, however, is that more creative people are finally willing to say publicly, yes, I use these tools.
242
00:17:59.964 --> 00:18:07.633
And that matters because it's almost impossible to have any serious conversation about creative quality when nobody will admit what they're actually doing.
243
00:18:08.274 --> 00:18:10.936
You can argue about ethics in the abstract forever.
244
00:18:10.996 --> 00:18:14.498
You can argue about policy and training data and labor impact.
245
00:18:14.838 --> 00:18:18.580
But creative industries eventually come back to the same brutally simple question.
246
00:18:19.341 --> 00:18:20.341
Is the work any good?
247
00:18:21.022 --> 00:18:22.583
Did it move you or surprise you?
248
00:18:22.643 --> 00:18:24.624
Does it feel authored?
249
00:18:24.904 --> 00:18:26.365
Is it worth your time?
250
00:18:26.505 --> 00:18:31.328
We haven't arrived at the mature version of this conversation yet because everybody's too afraid of being canceled.
251
00:18:32.008 --> 00:18:46.975
But it does feel like a useful first step because canceling people every time they admit to using a particular filmmaking tool just isn't helping us figure out where this technology belongs, where it doesn't, and what good work actually looks like.
252
00:18:47.736 --> 00:18:52.098
If people are finally willing to step into the light and say, yes, I used AI.
253
00:18:52.478 --> 00:18:53.758
Here's what I used it for.
254
00:18:54.159 --> 00:18:55.119
Here's what I made.
255
00:18:55.539 --> 00:18:57.120
Here was the economic impact.
256
00:18:57.420 --> 00:18:58.681
Here was the human impact.
257
00:18:58.981 --> 00:19:00.502
Here was an unintended benefit.
258
00:19:01.182 --> 00:19:02.183
then good.
259
00:19:02.703 --> 00:19:07.527
Because some of my favorite days in creative companies were spent arguing about what was good and what wasn't.
260
00:19:08.147 --> 00:19:09.408
Those were the good old days.
261
00:19:09.748 --> 00:19:13.271
And we can have them back, just as long as the conversation keeps evolving.
262
00:19:13.871 --> 00:19:15.873
So thank you for listening this week.
263
00:19:16.233 --> 00:19:17.174
I really appreciate it.
264
00:19:17.194 --> 00:19:24.399
We'll be back with our interview episode tomorrow with Ryan Black and Kite Jene from Luma Labs.
265
00:19:24.519 --> 00:19:28.742
Kite is a Ford deployed creative, which is the first time I've heard any company use that.
266
00:19:29.202 --> 00:19:42.845
But it's, you know, what Luma does when people are coming on and using its models, it teaches them how to use the nose base and agent systems by having creatives that will actually deploy into your company.
267
00:19:43.525 --> 00:19:44.585
So it's a really unique model.
268
00:19:44.645 --> 00:19:46.446
It's sort of stolen from the Palantir model.
269
00:19:47.786 --> 00:19:49.366
And, yeah, they're doing amazing things.
270
00:19:49.426 --> 00:19:52.247
Luma has one of the best models on the market.
271
00:19:52.887 --> 00:19:53.847
I think you'll really enjoy it.
272
00:19:53.867 --> 00:19:54.947
It's a really good conversation.
273
00:19:55.327 --> 00:19:57.628
Thank you for listening and tune in next week.
