In this second installment of the Cerebral Valley podcast series, co-hosts James Wilsterman and Max Child of Volley join Eric Newcomer for a thought-provoking conversation about the future of AI-generated voice and video — and what it means for our sense of reality.
From TikTok trends to the future of Hollywood and podcasting, the trio explores where generative video might take us over the next five years. Will AI content dominate our feeds? Will we even be able to tell the difference?
To put these predictions to the test, the hosts play The AI Video Turing Test — a game where they watch viral AI-generated videos from 2018 to today, ending with the latest clips made with Veo 3. Can they spot what’s real and what’s fake? And what makes some fakes feel too real?
This episode dives deep into the shifting boundaries between synthetic and human-made content — and makes one thing clear: we’re no longer at the bottom of the uncanny valley. AI is climbing fast.
The 2025 Cerebral Valley AI Summit will be held in London on June 25th
Timestamps:
02:50 — Predictions: AI in 5 years
15:00 — Hollywood perceptions of AI
18:55 — AI generated video games
23:05 — AI Video Turing Test
00:00:00
This episode is brought to you by Forethought.
00:00:03
Most companies build their customer experience in pieces.
00:00:06
Sales in one system, support in another, and onboarding
00:00:09
somewhere else. Forethought brings it all
00:00:11
together. Forethought is an AI system
00:00:14
made-up of advanced agents that handle sales, onboarding,
00:00:17
support, and retention. Each team manages its own
00:00:21
agents. The customer sees one unified
00:00:23
experience. Forethought powers over a
00:00:26
billion interactions every month for brands like Scale AI,
00:00:29
Cohere, Air Table, and Upwork. Learn more at Forethought dot
00:00:34
AI. Hey everybody, welcome to the
00:00:43
Cerebral Valley Podcast on the Newcomer newsletter.
00:00:46
We're getting ready for the Cerebral Valley AI Summit in
00:00:49
London on June 25th. Thrilled to be back with Max
00:00:53
Child. Hello, glad to be back.
00:00:55
And James? Wilsterman.
00:00:57
Hey, Eric. Good to see you.
00:00:59
We are talking about Max and James's favorite topic, voice
00:01:03
and the future of video. Though I guess you guys are
00:01:06
getting in. Both.
00:01:08
You got to get the branding right.
00:01:09
It's voice AI. OK, now voice AI and then video.
00:01:14
Eat it at you. I was like, I forget what we're
00:01:15
fighting about, but I was like, I noticed you add the AI to the
00:01:19
company. Yes, that's correct.
00:01:20
We were pioneers in the field of voice AI, and I've added myself
00:01:25
to the Wikipedia all right. We're in the Hayes Valley
00:01:29
Wikipedia. Do you know that?
00:01:30
I've seen this. Yes, yes.
00:01:32
OK, what? About our first accomplishments
00:01:35
clearly you know VO 3 is crazy like video creation has gone
00:01:40
insane. We want to cover a couple things
00:01:42
with this podcast where we think video is going and then we're
00:01:47
going to review some videos. We just had a lot of technical
00:01:51
difficulties trying to introduce AAI video Co host.
00:01:56
It seems like the problem was that fundamentally ChatGPT is
00:02:00
down because they introduce new models, which is honestly the
00:02:02
most disappointing failure mode for having a AI video Co host.
00:02:08
Since just ChatGPT working, it feels like it should be the
00:02:11
minimum. What are they spending the money
00:02:12
on? I don't know.
00:02:14
Apparently not server uptime. This has happened to us pretty
00:02:17
often by the way, we run a lot of our games on GPT API's and
00:02:20
it's, you know, every every hundred days or so you get, you
00:02:23
get nailed by this. They just dropped the prices,
00:02:25
right, 80% on O3, yeah. Does that affect you guys?
00:02:29
O3 is five times cheaper. I don't think we run any of our
00:02:31
games on O3 right now, but we do run our games on the real time
00:02:34
voice API, which is I think what we're here to talk about today.
00:02:38
The Her, the Samantha, Her model, as we called it a year
00:02:42
ago before Scarlett Johansson sued Open AI All.
00:02:46
Right. So a direct question.
00:02:48
Where do you think video is in five years?
00:02:52
We love predictions on this podcast.
00:02:55
And I think specifically I want you to get to, do you think
00:02:58
TikTok or movie studios like loom larger in this world?
00:03:03
And do you think AI is intertwined with human actors or
00:03:09
do you think we're pure AI movies?
00:03:13
There's a lot of layered questions there.
00:03:15
I think I would first say that anyone who's seen this sort of
00:03:17
viral Google VO3 model output, you know, on Twitter, which has
00:03:22
been very popular over the last week or two.
00:03:23
You know, there's a new viral sort of version of this video
00:03:27
content where it's a day in the life of a Star Wars stormtrooper
00:03:31
and they're making little joke logs of how they've crashed on
00:03:35
Tatooine and that kind of thing. And you know it, it looks, I
00:03:39
think as good or better than many scenes in certainly the
00:03:43
television versions of the Star Wars universe.
00:03:45
I guess you could argue maybe the film looks slightly better
00:03:47
if you're on a 70 foot screen in a theater.
00:03:50
But you know, it's it's close enough, in my opinion, to sort
00:03:53
of cut to the chase where you're like, this could be a part of
00:03:56
certainly like a Star Wars miniseries or something like
00:03:58
this. And so it seems hard to believe
00:04:01
that YouTube slash TikTok isn't going to take over this world.
00:04:05
I don't know how much of this type of content you'll see on
00:04:07
the movie theater screen. But I think what's more
00:04:09
interesting is like, given that you now have kind of on demand
00:04:13
video that's of the quality of, you know, Hollywood, Does
00:04:17
everyone get a personalized TikTok feed of their own AI
00:04:20
generated videos that are perfectly targeted at them and
00:04:24
and perfectly dispensed like their little, you know, their
00:04:26
little piece of candy that that they love.
00:04:29
I feel like that's one of the key for you're betting on long
00:04:32
tail custom video. I'm betting on ultra long tail.
00:04:35
I'm betting on like the barbell curve or what, you know, EU
00:04:38
curve, right Where on one end we'll have like monoculture,
00:04:41
we'll still probably have like, you know, the Taylor Swifts and
00:04:44
the sort of ultra popular films, you know, Barbie Oppenheimer or
00:04:47
whatever. And then most content
00:04:50
consumption will be super long tail, ultra personalized.
00:04:54
Everyone gets their own TikTok feed that's like AI generated
00:04:56
exactly for them. I think like, everything in the
00:04:59
middle is going to have a lot of trouble.
00:05:00
Like like our podcast, you know, until the middle.
00:05:04
Is done. Yeah, we're, we're probably in
00:05:06
the middle right now. So we need to get our AI Co
00:05:07
host. Taking over.
00:05:08
All right, James, what's your prediction here?
00:05:10
Well, I think that maybe podcasts will be safe for a
00:05:13
while. I think it seems hard to imagine
00:05:16
that we will have a big market for people listening to podcasts
00:05:21
purely hosted by AIS for a long time.
00:05:23
Because the reasoning through line isn't there just because
00:05:26
they're just gonna veer off. Yeah, I guess like I think
00:05:30
people mainly listen for the the hosts and the the sort of
00:05:34
personalities of the hosts and the relation to the real world.
00:05:38
You're like, we're so charming, how could it?
00:05:40
No, I don't know. I'm so charming.
00:05:42
If we're talking five years, I mean, I guess any, anything's
00:05:45
possible. Five years.
00:05:46
Anything's possible. The Silicon Valley point of
00:05:48
view, Yeah. I mean, I just think that
00:05:50
podcasts in general are just not that big of a business.
00:05:52
I mean, it's especially if we get self driving cars, I don't
00:05:54
think people are going to listen to that many podcasts in
00:05:56
general. I mean, I think you already see
00:05:58
video eating podcasts. Like, Oh yeah, yeah.
00:06:00
Well then I'm saying we're a. Video podcast.
00:06:02
I sure can. I don't think there's a
00:06:04
hardline. OK, well I guess I'm saying like
00:06:06
the concept of a talk show or whatever, right?
00:06:08
A long form talk show, which is sort of what James is talking
00:06:10
about I think. I guess I really disagree.
00:06:12
I think it will totally get eaten by AI generated content.
00:06:15
I listen to the Bill Simmons podcast.
00:06:16
I feel like the personal relationship I have with Bill
00:06:18
Simmons is, is of course, like, you know, meaningful because I
00:06:21
like that podcast more than other podcasts.
00:06:23
But I totally believe there's an AI out there that could create a
00:06:27
better show that I would like more.
00:06:28
And then I wouldn't care if it was a human or not.
00:06:30
Like there's nothing quintessentially human about the
00:06:33
Bill Simmons podcast that's like keeping me around.
00:06:35
I guess maybe maybe I'm underestimating Bill Simmons
00:06:37
abilities. It's possible.
00:06:39
Would you predict that the amount of content generated by
00:06:43
AI podcasts makes it uncommon to ever listen to a human generated
00:06:47
podcast in five years? That's what your sort of
00:06:49
prediction is, or would you go that far?
00:06:50
More airtime for human podcasts or AI podcasts in five years.
00:06:55
I say human. I'll say AI if we're including
00:06:57
all video like talk shows or whatever.
00:06:59
James, you're, you're on the hardcore human side, right?
00:07:02
You sure? My, my prediction is based on
00:07:04
like podcast players as they are today, like, you know, whether
00:07:08
that's YouTube or, or this. Is going with some kind of edge
00:07:13
gears here I think in Apple Podcasts and Overcast.
00:07:21
I do believe Max is on the right track here when we're thinking
00:07:24
about short form video, when we're thinking of whether even
00:07:28
Hollywood video will be mostly AI generated like anything
00:07:32
fictional where you're already looking for kind of a fantasy
00:07:36
type experience or world like, then obviously AI is just a
00:07:40
faster path to generate that. You can still have human written
00:07:43
scripts involved. I mean, sure, there will that
00:07:46
will be AI impacted as well. But yeah, I think that would be
00:07:49
where I would see this changing faster.
00:07:51
All right, Eric, what's your thoughts on this podcast
00:07:53
situation? I think what video hasn't had is
00:07:56
fast iteration. And so I think the biggest, I
00:07:59
mean, there are lots of things that are going to be interesting
00:08:01
about AI video, but on this sort of like long tail of videos, but
00:08:04
even popular videos like you get on TikTok, AI generates a short
00:08:08
video, sees how it how it's doing and then sort of dedicates
00:08:12
itself to videos that are getting heat.
00:08:14
Like I think this sort of just AB testing a video and seeing
00:08:18
what's working, what's interesting.
00:08:20
I think that's going to be a path for just like novel video
00:08:23
creation and is going to be really powerful.
00:08:26
So to me, I think TikTok and YouTube are much more important
00:08:30
than movie studios for video five years from now.
00:08:34
I also think the movie business, it's going to be IP wars.
00:08:38
You know, it's like we Studio Ghibli is like the top thing on
00:08:42
opening high. Like I just think that if you
00:08:44
actually try to make a movie that went to theaters about it,
00:08:47
you'd have a lawsuit. And I think the movie studios
00:08:51
with the actual IP have all these employees that are really
00:08:53
hostile to using the AI to generate their IP.
00:08:58
So I think largely YouTube, TikTok and this sort of user
00:09:03
generated type stuff is just going to be AI generated.
00:09:06
And so we're going to see a lot of the competition come there.
00:09:09
Yeah, That sort of gets into a question I have about the
00:09:13
future, which is given that the training data sets of all of
00:09:17
these AI is sort of where we are today or where we are up through
00:09:21
2020. And also given that, you know,
00:09:24
the only sort of rallying points for monoculture are sort of
00:09:27
nostalgia and IP. To your point, like are we sort
00:09:30
of in the like the end times for new content in some ways where
00:09:34
like forever going forward, like just the content universe of the
00:09:38
20 twenties will extend permanently into the future
00:09:41
because it'll be so deeply embedded in the training data of
00:09:45
the models that like sort of became, you know, became AGI or
00:09:49
whatever you want to call it that like, you know, the 1990s
00:09:52
and 2000s will be like weirdly important for the next 200
00:09:55
years. Because it's already pre
00:09:57
original. Content on the.
00:09:59
Pre AI, that the Internet had killed sort of generations or
00:10:03
that it was so self aware that you didn't have this.
00:10:06
Like we all dress this way until we all, you know, Yeah.
00:10:09
And so that banding was already sort of on the way out.
00:10:12
We see with new technologies does open up the opportunity for
00:10:15
new creative IP, right? I mean, Mr. Beast didn't exist
00:10:18
before YouTube and you know, Mr. Beast, the ultimate derivative.
00:10:22
He's doing The Hunger Games like.
00:10:26
I've kind of I've got it with Eric here.
00:10:27
I think Mr. Beast is gonna not not really last the test of
00:10:31
time, but. Sure.
00:10:32
But I think like obviously that that era of the Internet and
00:10:35
YouTube like created new stars that wouldn't have existed
00:10:39
without the Internet, right? So I guess I could imagine
00:10:42
something similar happening with AI where there's new types of
00:10:45
musicians, right, that are fully invested in AI.
00:10:48
There's new types of film makers and directors that are like go
00:10:51
100% in on AI, maybe new podcasts.
00:10:55
I mean, right now we're constrained in a couple ways,
00:10:58
like one VO3, what it's like 8 seconds at a time.
00:11:02
So people and humans end up having to do a lot of the work
00:11:06
because they're connecting all these different sequences if
00:11:09
they want to do it. So we're not even really seeing
00:11:11
true like AI creation right at scale.
00:11:16
And then the other thing, Sam Altman just wrote a new blog
00:11:20
post and I think he predicted that innovation or novel novelty
00:11:25
was like an upcoming feature of AI.
00:11:28
And so like in our five year window, I have a lot of optimism
00:11:31
that novelty could be coming. I think it's the big question.
00:11:35
And obviously if there's novelty, then there's new IP.
00:11:39
And then we're like, wow, they came up with the new Disney.
00:11:41
Like this is perfect. You came up with Pokémon in 20
00:11:44
seconds. Like that would then blow us up.
00:11:46
Then we're like every movie, it's like we get high on it.
00:11:49
I mean, we could even have this period where it's like we're so
00:11:52
in inundated with it and then we sort of pull back and we're
00:11:54
like, oh, we're fatigued of the AI movies.
00:11:57
That starts to ask the question of like, does it become a
00:12:00
monoculture or is super hyper targeted and everyone has a
00:12:04
different Pokémon that they're watching, right?
00:12:06
I at the end of the day, I know this makes me an old, but like
00:12:10
I, I, I watch TikTok, my wife watches TikTok.
00:12:12
We're lying in bed. We think we're so unique.
00:12:14
The TikTok comes up and it's like, oh, it understands me so
00:12:17
well. Then she gets the same one.
00:12:18
Like I, I think sometimes the Super niche tik Toks are like,
00:12:22
actually, no, there's a lot of common humanity.
00:12:24
Like we're all interested in the same things.
00:12:26
Like I, I think some of the long tail stuff is overrated.
00:12:30
Obviously we have hobbies and it's like, you know, chess or
00:12:33
League of Legends or whatever, so I'm not denying that, but I
00:12:36
think there's always going to be this urge for a shared culture.
00:12:39
Yeah, I think you're probably right.
00:12:41
I think the ultimate question is how many different human
00:12:44
interests really are there? And like especially ones with
00:12:47
any sort of density, like more than 1000 people share that
00:12:50
interest or more than 10 people share that interest.
00:12:52
And there's probably not that many, maybe on the order of
00:12:54
hundreds or thousands. So ultimately there might be
00:12:57
just a convergence around that and maybe the tail isn't that
00:13:00
long. But I do think like Drake
00:13:02
Apathy, who made like the original vibe coding tweet, had
00:13:05
an interesting tweet this week about VO3 where he said this
00:13:09
will be the first time in history that video is what he
00:13:12
called like directly optimizable.
00:13:13
Meaning like you could just run like millions and millions of
00:13:18
iterations on the same video, you know, and keep tweaking it
00:13:22
for almost no cost. Whereas historically tweaking
00:13:25
video has been sort of prohibitively expensive to, to
00:13:28
do more than a few variations or more than a few kind of
00:13:30
iterations and tests. And so video is going to become
00:13:34
a lot more like, you know, text or something like that, where
00:13:37
you can iterate and test it so rapidly that you can converge on
00:13:41
sort of the like optimal solution for a given problem,
00:13:44
even if like the optimal solution is just like a very
00:13:47
particular version of a meme or whatever.
00:13:49
And so it'll be interesting to see what those shared pieces of
00:13:52
content look like in the future and what the sort of hyper
00:13:55
optimized version of something that that was, you know, merely
00:13:58
popular when we were he's. Saying it's cheaper, not yet
00:14:01
like there's going to be a lot of costs.
00:14:03
Like I do think another reason to be bearish on hyper
00:14:07
individualized video in the next 5 years is just everybody's
00:14:11
like, oh, wave away costs, wave away costs.
00:14:13
Actually, it's like make one expensive AI video for everybody
00:14:18
in the beginning and then once we get costs down, you know?
00:14:21
Yeah, for sure. I do think five years is such a
00:14:24
long time in this AI era that we live through though, that it's
00:14:28
really hard to predict where costs end up for anything.
00:14:31
I mean, like we just said, Open AI reduced costs for O3 today,
00:14:34
80% down from what it was yesterday.
00:14:37
So I think things just change dramatically.
00:14:40
I think there will be a successful Pixar style AI
00:14:45
company. Like I think if we look at Mid
00:14:47
Journey, part of why Mid Journey has been successful is it did
00:14:49
sort of it's like, OK, we like this design aesthetic.
00:14:52
They're sort of playing kingmaker on certain looks to
00:14:54
make it work. And so I think there will be
00:14:57
sort of like AI studios basically trying to make movies.
00:15:00
Given that your your wife is in the filmmaking industry, like
00:15:04
what is your sense on is Hollywood going to adopt these
00:15:09
tools like for, you know, special effects or for like
00:15:12
certain shots or to like edit certain things?
00:15:14
Like how much blowback is there against it versus how much
00:15:17
excitement? My life is in documentary.
00:15:20
I think she's a little more open to it, given what I do.
00:15:23
But certainly in those circles you can feel like they're way,
00:15:26
you know, go to their parties. They're way more hostile to AI
00:15:29
than be unfathomable to us. So yeah, there's definitely a
00:15:33
lot of hostility in Hollywood. I mean, I, Tom Doton has been
00:15:37
doing some reporting for newcomer and I don't know my my
00:15:40
sense from his story is that it's like Hollywood secretly
00:15:44
the, you know, the execs are gonna use it and it's gonna
00:15:46
happen and and there will be this sort of clash with the rank
00:15:49
and file. I mean, documentary, given it's
00:15:52
about the real world, you know, has its own reasons to be sort
00:15:56
of not as AI generated, but it'll continue to create sort of
00:15:59
this blur between real and fiction.
00:16:01
And I mean, we kind of talked about this earlier, but like, I
00:16:04
already think a lot of this AI video looks more real than like
00:16:07
an effect shot in a Marvel movie.
00:16:09
And I think Hollywood obviously moved in the direction from
00:16:12
practical effects in the, you know, seventies, 80s and 90s to
00:16:15
100% CGI for the most popular movies today.
00:16:18
And I just think that it's sort of inconceivable that you
00:16:21
wouldn't take this technology that lets you create at least
00:16:25
like effect shots for like 1-1 thousandth as much money, like
00:16:29
1100 thousandth as much money and like probably looks better
00:16:34
already in 20. 25 to control the camera right?
00:16:37
Like. There are people working on that
00:16:38
for sure. We're really close, I guess.
00:16:41
I have a question. I just think that we're dancing
00:16:42
around that we haven't gotten to you, which is just like we're
00:16:46
predicting that all this video is going to be generated with
00:16:49
AI. We're predicting that Hollywood
00:16:50
studios will adopt it for special effects.
00:16:54
I'm still kind of unclear what will the human role in any video
00:16:58
production be? Will it be writing?
00:17:00
Will it be acting? You know, someone's likeness,
00:17:03
their voice? Will there be anywhere?
00:17:05
Humans part of the process. I think storyboarding will.
00:17:09
I mean, I think the the script might hold on the longest or
00:17:14
like ensuring coherence across the script.
00:17:18
I think that's where I'm at. James and I have been making a
00:17:21
game, and definitely today the script is the thing that is the
00:17:25
hardest to get the AI to do a good job on.
00:17:27
Like we're doing an opening cinematic scene for our game and
00:17:30
James and our content team have devoted weeks and weeks and
00:17:33
weeks to iterating on the script and reshooting the script and
00:17:35
everything. So it does feel like the current
00:17:37
thing that's not good. But you know, if you throw the
00:17:40
script into Claude 4 or 03 and say, hey, like, clean up the
00:17:44
script and like, give me a better version of it, like it
00:17:47
gives pretty good notes and pretty good feedback.
00:17:49
So maybe it can't hold a coherent script across 120 pages
00:17:53
today, but it's doing pretty well at a six page script right
00:17:56
now in terms of at least like editing and understanding the
00:17:59
narrative arc and everything. So I just and does it have like
00:18:02
a target offs and stuff later and like the sort of classic
00:18:05
stuff? You can ask it for pay offs.
00:18:07
I mean, yeah. It's kind of like you, You as a
00:18:09
human have to know what you're looking for, and then it gives
00:18:12
you good ideas for these things. But yeah, it can't do everything
00:18:15
itself. But I do think that's key to it.
00:18:18
Like you're like, oh, does it have a key, the key element of
00:18:20
story that like there's some emotional payoff or something.
00:18:24
You're like, oh, hand wave that away.
00:18:26
The humans will make sure like that's that's what I'm saying.
00:18:28
So which? Is why short form video maybe?
00:18:30
You're saying it's like a script producer or something?
00:18:33
Or it's like the editor of the script rather than the writing.
00:18:36
I don't know, a year ago it was really, really effing bad at
00:18:39
this. And now it's like getting kind
00:18:41
of good at it. And so you're like, I see where
00:18:43
the curve is headed on this. I don't know.
00:18:45
I think, I think a year from today it could be pretty good.
00:18:48
I have a prediction that within five years video games will be
00:18:53
completely AI generated at the pixel level, effectively using
00:18:58
video generation models that are interactive as opposed to
00:19:02
current strategies to render in a game engine 3D environments.
00:19:06
Just for context, the way most games are made today is inside a
00:19:09
game engine. You put a bunch of shapes in an
00:19:11
environment and walls and light sources, and then you try to
00:19:14
figure out how the light bounces off various things and what
00:19:17
camera the player's looking out of and such.
00:19:21
You're saying the future is we're basically just gonna take
00:19:24
VO 3 but like, it generates the whole environment in real time
00:19:28
all the time. And so instead of watching a
00:19:30
little 8 second clip of something funny, you can walk
00:19:34
around the world that VO 3 has generated for you and it'll all
00:19:36
just change as you're walking. Around doing this with Minecraft
00:19:40
already right? I mean, I've seen weird demos of
00:19:42
it, but it doesn't seem real yet.
00:19:43
Yeah, they're, they're, they're doing demos where there's
00:19:47
Minecraft levels being generated in front of you as.
00:19:50
There's anything we know about demos that every tech demo that
00:19:53
has ever been given has turned into a real product that someone
00:19:56
can deliver. I don't think it'll replace
00:19:58
everything. I think this is the classic
00:20:00
like. All or nothing, false dichotomy.
00:20:03
I mean, I think it'll be interesting.
00:20:04
Doesn't seem like the Switch just came out first of all, like
00:20:07
Nintendo is like building games on like 5 generations old
00:20:11
technology. It seems like it's not even like
00:20:13
OLED on the device. So I just think you're like
00:20:16
overestimating how quickly an industry that's not AI will come
00:20:20
for it like new. I think new companies
00:20:23
fundamentally and new studios will have to create these games
00:20:25
and they will have to disrupt the existing studios.
00:20:28
I don't think the existing studios will necessarily embrace
00:20:31
it quickly. And the existing studios have IP
00:20:34
and so it's going to be a slow transition.
00:20:36
Five years is really fast for this kind of thing.
00:20:38
To Eric's point, just from like a business model and technology
00:20:41
perspective, I also just think that I've seen all those demos
00:20:43
that you've seen of people generating Minecraft with video,
00:20:45
and like, it's not clear to me what they're better at, right?
00:20:48
OK. They're more expensive, they're
00:20:50
slower, the world looks like crap, the walls change all the
00:20:53
time. Like, you know, there's no other
00:20:55
players in there. There's no game mechanics like
00:20:57
they're worse at everything, right?
00:20:59
OK, that doesn't mean someday they won't be better at
00:21:00
something, but I've never like heard the pitch for why it's a
00:21:05
better way to build a game world in.
00:21:07
You can do anything. You could do anything cheaper.
00:21:09
You can prototype everything you know in seconds, you can have
00:21:13
any game you want, right? But.
00:21:14
You can do anything in a game engine.
00:21:15
You can drop whatever shape you want in the game engine.
00:21:17
You can make any character Sprite.
00:21:18
I mean, maybe it's easier and faster someday, you know, but
00:21:21
it's also extremely non deterministic, right?
00:21:24
Like, you know you, you're like, hey, I want a wall there.
00:21:27
And then sometimes the wall moves or whatever.
00:21:29
I don't know. Again, I believe there is a
00:21:30
world in which they're better at some things, but I've never
00:21:33
heard which game you would want to build in this thing, even as
00:21:35
sort of an imaginary awesome version of it.
00:21:38
It might come back to this like long tail argument where you can
00:21:40
make custom worlds for every player and somehow that's more
00:21:42
interesting than, you know, a more structured game world that
00:21:45
you know, everyone was in the same one, I mean.
00:21:47
Fortnite is a huge game. Even if there is one game that
00:21:51
is made this way, that is like, sure it's the everything game
00:21:54
because it changes, you know, as you develop.
00:21:57
As as you go through it through. It you know, but that's a much
00:22:00
smaller claim. There will be 1 super intense
00:22:03
game where you can do this. Or maybe there will be a whole
00:22:05
category? I just think if it's technically
00:22:07
possible, well, it seems like it will be better than waiting, you
00:22:11
know, five years to build a new GTA or something, that it'll
00:22:15
just be kind of laughable that you would invest in a studio to
00:22:19
do that. I don't fully agree, but one
00:22:21
thing I will say is like, and this isn't games purely, but if
00:22:25
you like think about Harry Potter, right?
00:22:26
And they're psyching themselves up for this like series on Harry
00:22:30
Potter. But there's so many assets fans
00:22:33
have already created so much. It feels like given a little bit
00:22:36
of AI juice, like the fans are ready just to create their
00:22:40
preferred version of Harry Potter and all the like.
00:22:42
Fandom can fight with itself about exactly like how everybody
00:22:46
should be portrayed. And I don't know, I think that's
00:22:49
almost more exciting. I think this Harry Potter series
00:22:51
is coming at the worst first time.
00:22:53
Like this is the time where the fans are about to be empowered
00:22:56
to really make a full length series themselves.
00:22:59
And I'm super excited about that type of thing.
00:23:02
Let's let's dig into some actual real videos instead of just
00:23:07
imagining the world that could be, the world that is the insane
00:23:11
world that is. Let's pull up some videos and
00:23:13
get your reactions. All right, we're watching an AI
00:23:20
video from seven years ago. You won't believe what Obama
00:23:25
says in this video, Wink. We're entering an era in which
00:23:31
our enemies can make it look like anyone is saying anything
00:23:34
at any point in time, even if they would never say those
00:23:37
things. Moving forward, we need to be
00:23:39
more vigilant with what we trust from the Internet.
00:23:42
That's a time when we need to rely on trusted news sources.
00:23:46
My first reaction was this was better than I thought it would
00:23:48
be. Better than I thought it would
00:23:49
be from 2018. Yeah.
00:23:51
It had that effect of the the mouth moving off to the side
00:23:54
because it basically was like someone cut a hole in Obama's
00:23:56
mouth and then just like put someone else's mouth in the
00:23:58
video. I think that's kind.
00:24:00
Of what you figure out, it's a voice actor.
00:24:02
You're like, OK, that's not AI so.
00:24:04
So the TTS voice wasn't there. I mean, it was, it was pretty
00:24:06
good. It comes back to our point about
00:24:08
the the monoculture versus the long tail here, which is Obama
00:24:12
obviously very monoculture figure It was worth the
00:24:15
investment to cut a little hole in Obama's face and have him
00:24:18
make silly things because you knew that was going to go viral.
00:24:21
Whereas in the modern era we can.
00:24:23
Spend a ton of money on this video.
00:24:25
More than more than you would think, probably.
00:24:28
All right, Next. Oh, that's hot.
00:24:35
That's hot. Uncle Phil, come try this fresh
00:24:40
pasta of belly. Why do we think that this
00:24:47
particular Will Smith eating pasta video went so viral in
00:24:50
2023? I know now it seems like who
00:24:53
cares? I mean, this is definitely the
00:24:54
argument for like. Wow, things have really
00:24:57
improved. This is a horrifying and
00:24:58
disgusting video and I think people all enjoyed sharing that.
00:25:02
Look how terrible AI is. I feel like there might have
00:25:04
been an element of dunking on how bad so.
00:25:07
This Vadnais is what's useful about it, because it suggests
00:25:10
OAI is not going anywhere. All right now we're moving on
00:25:14
from old BuzzFeed videos in the Stable Diffusion era not too
00:25:18
long ago to pretty recent history.
00:25:21
Sora. Yeah, so there's two dogs
00:25:33
podcasting on what looks like a an Alpine hillside or something
00:25:36
like that. That one looks pretty realistic.
00:25:40
I mean, I would say like, I don't think it passes the Turing
00:25:43
test as it were, where I'm like, that looks like a real video.
00:25:46
There's just enough issues with it.
00:25:48
But you know, it's it's starting to get into the era of like, oh,
00:25:52
this kind of looks like, you know, a realistic image, I
00:25:55
think. Whereas obviously what we saw
00:25:57
before is nowhere close to that. What's interesting to me else is
00:26:00
that they had these watermarks on there, which I think now
00:26:03
they've kind of allowed you to remove the watermark, but I
00:26:07
think there there was this fear that these things would be so
00:26:10
real that you really need a persistent watermark on all
00:26:14
these videos. They're more real, we careless.
00:26:16
Yes, that's what's surprising to me.
00:26:18
It's just like, or we've become more comfortable with the idea
00:26:22
that any video on the Internet could be AI generated and
00:26:26
everyone kind of knows that, but I'm not sure that's true.
00:26:29
I think it's a race to the bottom.
00:26:30
I think it's just there's enough competition in AI that these
00:26:32
companies know that they want to valuable.
00:26:34
They need to provide stuff without their label on it.
00:26:37
Be somebody else is all right. What's next?
00:26:40
Welcome to a nonexistent car show.
00:26:42
Let's see some opinions. I mean man, the acceleration is
00:26:47
crazy, you look far, step on the pedal and you are there.
00:26:52
I feel safe with him in an SUV and it seems to be like the
00:26:56
right type of car for him. I think the range is only only
00:27:01
going to get better. Sorry.
00:27:04
We don't want to drive gas cars anymore, Yeah.
00:27:07
No more gas cars. You can say I'm kind of a kind
00:27:12
of a misfit here, but don't tell anyone.
00:27:15
I've just bought an electric car.
00:27:17
I think it's really great for families and for little babies
00:27:20
with all the safety features that these SUV's have.
00:27:23
But what you're really seeing is that technology is going to be
00:27:27
very, very important in terms of how we go forward.
00:27:30
It was great to come to the conference because my husband
00:27:36
loves cars. I think I have to buy an EV now.
00:27:40
I love my muscle cars but I try to stay as healthy as I can so I
00:27:46
can make it to the next car show.
00:27:51
I mean, as someone who's watched a lot of like random YouTube
00:27:54
videos of people looking at cars, it's pretty damn good.
00:27:58
I mean, I would say it's it's like trained on YouTube videos
00:28:02
of people looking at we. Can tell it's fake.
00:28:04
Yeah, boomers maybe can't. I don't.
00:28:07
There's only some humans who'd get confused.
00:28:09
I think if you presented this to me and I didn't know it was
00:28:11
fake, I would have to think about it for a bit.
00:28:14
No, you're. Like no, I would immediately
00:28:16
know. Certain certain clips.
00:28:17
Certain clips. Certain clips are a little more
00:28:20
obvious than others. So if you if you presented the
00:28:22
whole video, I might start to think it's fake.
00:28:24
But if you presented anyone clip like I think I I'd be pretty hit
00:28:28
or miss whether I know it's fake.
00:28:30
Sure, the stills are a little harder.
00:28:32
It's sometimes the movement that gives it away.
00:28:34
I do think like something we haven't really reflected on is
00:28:37
how Sora got announced, I think over a year ago and like barely
00:28:42
launched and has just been obliterated by VO.
00:28:45
I mean, isn't that like a, an interesting story that they're,
00:28:49
you know, I think what we would all argue that GBT is basically
00:28:52
best in class or, you know, very near best in class at all times.
00:28:55
And I feel like Sora was supposed to be the best and
00:28:59
Google has just come through with the steam.
00:29:02
Trains. Dominant, but it's crazy like I
00:29:05
mean I don't think any of us would have predicted a year ago
00:29:07
oh, Google's video model will be by far the best video model a
00:29:11
year from today like it won't even be close it's not like it's
00:29:13
like oh head to head right now but.
00:29:15
I was a defender of Google. But why do we think they're so
00:29:18
particularly good at video? Is it because they own YouTube?
00:29:21
Yeah, I mean, they're just like doing YouTube.
00:29:24
Yeah, but like everybody else stole all of the YouTube videos
00:29:27
to train their models, what I understand from the buzz in the
00:29:30
industry. But like, why are they still so
00:29:33
much better at this than everyone else?
00:29:34
It's it's sort of an interesting question to me.
00:29:36
Like does Sam not care? Does Dario not care?
00:29:39
Or is there something else about it?
00:29:41
My theory would partially be that everyone's a little compute
00:29:45
constrained and they just have different priorities and video
00:29:49
is not going to be open as priority.
00:29:52
Like they are more worried about anthropic and coding models and
00:29:55
competing and things like that. But I don't know, just a theory.
00:29:58
So do you think they'll maintain their advantage, Google?
00:30:01
I mean, that would sort of be the case for that.
00:30:03
Yeah, I kind of think that Google will be a little bit less
00:30:06
compute constrained and have all this YouTube data and seems to
00:30:10
care more about this. So yeah, I think they'll
00:30:12
maintain the advantage for the next year or more.
00:30:15
All right, one more. Kangaroo the airport.
00:30:17
All right, let's. I actually haven't seen it.
00:30:19
All right, we're going to watch. I've never seen this before.
00:30:22
This is sort of an infamous one, that kangaroo in the airport.
00:30:35
I mean, the kangaroo is high quality, but maybe, you know, as
00:30:38
human beings, I'm not as good at the animal, you know, But the
00:30:42
lips on one of the women, you can sort of tell, but it's
00:30:45
getting very picky. Well, there's a couple of things
00:30:47
I noticed. There's this very good kind of
00:30:50
comedic zoom in effect, right? And that was probably prompted,
00:30:55
which is pretty interesting. And then I don't even know if
00:30:57
this is a real language. I don't think it is is that.
00:31:00
Do you guys the audio is weird? Yeah, did they strange?
00:31:04
Did they prompt it to create a fake language also?
00:31:07
I feel like this fits into it was a Maxis theory that just
00:31:11
like this is the end of culture sort of thing.
00:31:13
It's like just the cinematic language.
00:31:15
It's like the zoom in is just like whatever it's like.
00:31:19
Exactly. It's the exactly like that.
00:31:21
And so it's just like the video language we had up until now.
00:31:25
We're just gonna like. Roll with it.
00:31:26
Other than me knowing that Kangaroos are very unlikely to
00:31:29
be in the airport, I would be easily be fooled.
00:31:31
I mean. And the language piece,
00:31:32
obviously, but like, like if there was, you know, no audio,
00:31:35
I'd be like, the only reason that doesn't seem believable to
00:31:38
me is because I've no Kangaroos aren't supposed to be in
00:31:40
airports. Like there's nothing about the
00:31:42
video to me that. Stood oh man the comments yeah
00:31:44
I'm definitely getting scammed at 80 like you're probably
00:31:50
gonna. Get scammed at 40.
00:31:52
Like, and some of them are just, he's just sitting there all
00:31:54
polite. I mean, the, the other
00:31:56
terrifying thing about this is like, there are plenty of people
00:31:59
who don't really care whether it's real or not, right?
00:32:01
That's been a constant Internet problem.
00:32:03
We get obsessed with like, can they tell?
00:32:05
But they, there are people, there are, there's a whole genre
00:32:08
of fake videos where people are like, who cares if they're
00:32:10
acting or not? It's a little nice sketch, you
00:32:12
know, like this. Is why I'm saying from the
00:32:14
beginning of this podcast that people are going to be totally
00:32:17
cool with the vast majority of their tick tock feed being fake.
00:32:20
I, I mean, you guys are well, like now, man, like there's all
00:32:23
about the emotional connection or whatever and I just.
00:32:25
Don't believe I I believe you when it comes to short form
00:32:28
video. I when you're talking about like
00:32:30
2 hour long podcasts, I'm just a little more skeptical.
00:32:33
And podcasts, not to bring this to a really dark tone, but like
00:32:37
with the protests in LA, you're seeing like the creation of
00:32:40
almost like two different stories about how serious it is
00:32:44
with images. Now, obviously I don't think
00:32:46
those are fake images. Some of them are images from
00:32:49
different riots and protests, and not necessarily the current
00:32:52
one, but we've already seen this tendency that you can have a lot
00:32:55
of influence on the Internet by showing images that tell the
00:32:59
story you want. And now, obviously with AI,
00:33:02
we're headed down this world where people have a strong
00:33:04
desire to see the world they want and we'll be able to
00:33:07
deliver it to them. Yeah.
00:33:09
I mean, it is an interesting question as to whether or not
00:33:11
like traditional news brands, you know, gain or lose esteem or
00:33:15
status because like the assumption would be anything you
00:33:18
see on whatever CNN is real right now.
00:33:21
Obviously some people would debate, you know, whether or not
00:33:23
they framed it correctly or whether the hosts are all biased
00:33:25
or whatever. But like most people believe the
00:33:27
footage they see on CNN is real, myself and.
00:33:29
Right. Interesting.
00:33:30
There'll be no AI brands, brands that are like, we don't have it,
00:33:34
no. AI exactly like like a news
00:33:36
brand or even a movie brand as we discussed earlier, I believe
00:33:39
there is probably room in the market for saying we nothing is
00:33:42
ever AI that we show, we verify every clip.
00:33:44
It's kind of like what you do with investigative journalism
00:33:46
where your brand is I only write the truth, right?
00:33:49
You know, I will never put rumors.
00:33:52
I will never put things I haven't verified right.
00:33:54
And I think that is compelling as as an angle I.
00:33:56
Know it's exhausting it'd be so much more fun to be one of these
00:33:59
people just says whatever the fuck like yeah trying to be true
00:34:03
in a world where people can say whatever they want it's.
00:34:05
I kind of think that hardware companies, like the phone
00:34:08
companies are going to have to create new technology that says,
00:34:12
hey, this video that you took on your phone is provably a real
00:34:17
video. And we anyone can go check that
00:34:21
it's a real video. And maybe the social networks
00:34:23
will have some auto checking that it's a real video.
00:34:26
But I think it almost has to go down to the hardware level to be
00:34:29
able to prove that. We saw an interview with Demos
00:34:31
to Sabas from who runs Google DeepMind and then also Sergey
00:34:35
Brandon, which was interesting where Demos said, and I didn't
00:34:38
know this if they'd revealed this before that there is a
00:34:40
secret watermark in every VO3 video, you know, so they can
00:34:43
tell every every video that's fake.
00:34:45
He sort of alluded to that there was a collusion scheme that the
00:34:49
major players all have watermarks in their videos and
00:34:51
they all know each others watermarks kind of that.
00:34:53
So they know what a sort of video looks like or they know
00:34:55
what a, you know, whomever. And that was very interesting
00:34:58
because it would make me wonder if you have like a deep mind
00:35:01
situation where someone decides to like be the only AI video
00:35:04
company that doesn't put a watermark in their video, you
00:35:06
know? Right, well, xai would be the
00:35:07
class. Yeah, they're, yeah, they're the
00:35:09
only ones who can get through the filters.
00:35:10
And so everything on YouTube is real.
00:35:12
But then there's this sort of like vector attack from, you
00:35:15
know, you know, XAI or DeepMind or whatever to there's to upload
00:35:19
a bunch of. Yeah.
00:35:20
Yeah, yeah, yeah. Anyway, it was just kind of
00:35:22
interesting. I didn't realize that they were
00:35:23
so serious about that. I mean, I think a recurring
00:35:26
theme, I've been optimistic that Chachi, BT and other models are
00:35:32
going to bring culture into more of a shared reality because
00:35:35
instead of going to a bunch of different links, you have a
00:35:38
model that's trying to get a vision of reality and it's at
00:35:40
least delivering one reality. And if the model gets it wrong,
00:35:43
maybe all of society is misled. But at least we like are getting
00:35:47
back to a shared culture. I feel like this video, the
00:35:51
experiment feels sort of the opposite.
00:35:53
I mean, I think partially because it's going to be
00:35:55
narrowly tailored. I don't know, I it feels like I
00:35:58
worry about video being a path to manipulate the culture in a
00:36:02
way that I feel like the text models, I feel all more
00:36:05
optimistic about them and forming the culture.
00:36:09
I agree with you, but maybe with a slightly different angle,
00:36:11
which is that I think fundamentally video is like by
00:36:13
far the most effective way to convey information to people,
00:36:16
right? I mean, I think that like people
00:36:19
take in, you know, the whole, you know, pictures worth 1000
00:36:22
words, maybe a video's worth 10 words or whatever, like.
00:36:25
Right. I guess it's dangerous, yeah.
00:36:27
Yeah, exactly. They take in the information
00:36:29
from video at such a higher bandwidth level than they do
00:36:32
from books. You know that I think that it's
00:36:35
a really, really powerful communication and teaching
00:36:39
medium. And so in some sense, like
00:36:41
everything that's on video, yeah, and propaganda, everything
00:36:44
that's on video is just supercharged in terms of its
00:36:46
ability to influence people and seem like it's reality, right.
00:36:50
So I sort of think that, you know, the title of this episode,
00:36:52
I think was like the war on reality.
00:36:54
I think the war on video reality is sort of the most important
00:36:58
one in some ways because it is the highest bandwidth, I think,
00:37:01
communication medium that we have if.
00:37:03
It's in video. People believe it.
00:37:04
Well, I think your point, Eric, is that people go to Chachi BT
00:37:08
to kind of learn about the reality of the world.
00:37:10
They ask questions, questions about things, they ask for
00:37:13
advice, life advice that they're actually going to put into
00:37:15
practice. I think, you know, people go
00:37:18
scroll TikTok for a different reason.
00:37:20
It's really to just zone out, be entertained.
00:37:22
People watch Netflix to be kind of in these fictional fantasy
00:37:25
worlds. I just think these are very
00:37:27
different use cases. Text, first video.
00:37:29
And I think that's what you're you're noticing.
00:37:31
Preparing you to go out into the world and sort of like, livid.
00:37:36
Whereas I feel like video, you are like living the world.
00:37:39
You're like, oh, it's as if you were like living.
00:37:41
And you're like, take what you've seen in video and you're
00:37:43
like, that's the world. I've seen it.
00:37:45
You know, I don't know. It's, yeah.
00:37:48
It seems dangerous, but very compelling.
00:37:51
Anyway, that's our show. Thanks for sticking around.
00:37:54
We're getting ready for this Rebel Valley AI Summit, London,
00:37:57
June 25th. Thanks for joining us.
00:38:30
Yeah, I'm still eating the spaghetti in 2025.