Creative Flux

Pierson Marks (00:00)
Hey, episode 21.

Bilal Tahir (00:00)
Hello? Hello, hello?

I was just going through some of our previous episodes and I didn't even know I did this, but I will say hello, hello apparently that's like...

Pierson Marks (00:09)
Really? I don't even notice. I don't even notice. Yeah, it's

fun. mean, I can't believe we've been doing this now for like, okay, 21 episodes, four episodes a week. That's like, it's a solid, a solid number.

Bilal Tahir (00:21)
yeah what we started when like july july july probably june july yeah

Pierson Marks (00:24)
Five months? Almost. We've been covering

the space. It's cool. I know we go off track sometimes on the general media stuff, which is fine. We just talk about everything. AI image. So if you're listening to this for the first time, we talk about AI images, video, world models, audio, music, anything in the creative sphere. But we go on tangents. Sometimes we talk about cryptocurrency. Sometimes we talk about agents. Sometimes we talk about random stuff. ⁓

Bilal Tahir (00:51)
Mm-hmm.

Pierson Marks (00:52)
But it's always fun. It's always a journey. We never know where we'll start. We normally know where we'll start, but we never know where we'll end up.

Bilal Tahir (00:59)
Yeah,

yeah, that's the beauty of going down rabbit holes. you learn so much, feel like, doing that. So what do we have on the menu today?

Pierson Marks (01:02)
⁓

Right. Totally.

Well, what

do we have? think, so we were just talking about this right before the call, creative writing. I want to touch on this real quick because we talk a lot about images, video, like generative media. But one part of the media is writing. And so there was a model that came out by a Chinese lab, Moonshot AI, right?

Bilal Tahir (01:15)
Yes.

Yeah.

Right, Moonshot AI. They're known for the Kimi K2. And Kimi K2 came out a few months ago and was pretty good. And the writing was really good. Like it was really good at creative writing. And most recently, last week, they released Kimi K2 Thinking, which is the thinking version of that model. And it jumped to, think, number. It beats, basically, GPT-5 on Humanities last exam, which is a pretty big benchmark for how smart a model is. Basically broke all the benchmarks.

it only cost apparently $4.9 million to do the whole training run. So it was actually, it's weird.

Pierson Marks (02:00)
Right. Do you know the details on that though?

Like is it 4.9 in the final training run? Like what's the...

Bilal Tahir (02:05)
That's

always a thing, it's funny because it harkens back to the deep-seek moment we had and everyone freaked out. Maybe we were used to it, but this was actually as big of a deep-seek moment because they did just break all these benchmarks. And the model is a fraction of the cost to serve.

⁓ Compared to other models. So it was a ⁓ kind of a blow for all the big labs in the US, you know, who Don't want to make sure their models have a premium on them, right, which is kind of gone away. So I for some reason Kimi K into hasn't I feel like, you know, Escape the Twitter bubble yet, but it is very interesting and very cool and very different like you were saying the creative writing is very different on it for some reason I think and I think it's because of the training data they use it doesn't sound like

other LLMs, if you've used other LLMs, you've noticed that they all kind of say the same thing. It's not just X, it's Y, like, you blah, blah, blah. And KimiK2, at least in my experiments too, it was very different. Also, we'll say fuck or whatever, like, you know, which, ⁓ yeah. Yeah, yeah, I I'll read out. I did this one experiment where it, and it's funny, because the... ⁓

Pierson Marks (03:00)
Really? so it will ⁓ occur.

took your tweets

and then you try to make a tweet like below.

Bilal Tahir (03:12)
I just like,

was like, just give me a hard hitting paragraph about, you know, just someone, you know, who needs to hear the bitter truth. And this is what it tells me. Nobody tells you that the heaviest thing you'll ever carry isn't debt or disappointment. It's the weight of the person you could have been standing right behind you, whispering in your ear every single day. This isn't how it was supposed to go. And you can't put that down. You can't set it on a shelf and walk away. That fucker lives in your skin. And this tweet actually was...

Pierson Marks (03:36)
⁓ Did you say to

be explicit or did it like come out explicit? Did you say or no?

Bilal Tahir (03:40)
Yeah, hit fuck.

It's hit fuck. And actually, it's funny. I put it, I tweeted it out. And one of the first people to like this was the co-founder of YouTube, which I'm like, wow, that's a random like, you know, so.

Pierson Marks (03:49)
But that's very hard. Wait,

wait, wait, just to clarify, did you explicitly say be explicit or did it just do that without prompting?

Bilal Tahir (03:58)
I think I was tweeting, like I was chatting with it and it was giving me, I was like, just make it more fucking like, you know, like raw. And so I did use the word fuck before, but so it wasn't totally unproper, but it didn't say, ⁓ you know, I can't see blah, blah, blah. You know, it just did it. So really cool.

Pierson Marks (04:12)
Arrrr

Right. so

I'm curious. So obviously, creativity and writing is very qualitative. And it's hard to measure what makes writing good. Because if there was a benchmark, like what makes writing good, it would be easier to test for. Was there something out there that is there a benchmark that helps to do that? Is it like kind of an LLM arena type thing where you put ⁓ Kimi K2 on the left and Chachi BT and you have preference?

Bilal Tahir (04:28)
Right.

I

think there is a creative benchmark, but I honestly don't know how you would rank it. I other than maybe users kind of selecting like LLM arena style, like which paragraph sounds better, but because you're right. I mean, the creative writing is like very subjective. What's Hemingway to one person is like, you

grows to another person. So it almost is so context dependent. ⁓ So it's very hard to measure. It's the same thing with Jalpo. I mean, we try to make our podcast transcripts sound amazing. But again, we have customers who want to have professional podcasts to just shooting this shit, like have a fun podcast. And so how do you kind of capture both of those things in one?

Pierson Marks (05:04)
Totally.

I mean, and

it's hard. mean, if you looked at our code, the prompts that we use to generate scripts is pretty in depth and they have like these weird...

They have cases where if this, then that. And we just adjust that prompt based on certain things. And it's hard. We spend a lot of time prompting to get the current models that we use to a place that produces an engaging script that doesn't sound AI. And I think we actually do a pretty good job. think people at Common said that the scripts are actually generated really well for realism. The thing that, though, I'm most curious about, and it goes back to this overall theme about models.

Bilal Tahir (05:52)
Right.

Pierson Marks (05:58)
always has been a debate between one model to rule them all versus smaller, more specialized models. And I think I'm pretty much now convinced, now more than ever, a few months ago I may not have been, I might have been one model to rule them all, but I'm more so convinced that there's going to be a lot of different models because I think you have to make trade-offs. I think that's the thing where

Bilal Tahir (06:16)
Hmm.

Pierson Marks (06:21)
you go and you, like most people will use ChatShipT. Most people will go use Gemini or Anthropics, like Claude or Grok. And like you'll use those big model providers and they each kind of have a different taste. Like why does somebody go to Grok on X? Because it's a little bit more raw. It has access to Twitter, but that's not model level. But if you're looking for information and knowledge-shaking, you don't really care about the output, you just want like the truth. Maybe you could have one model rule at all.

Gemini could be that person or that model. Chachi Buti could be that person and model. But when you come down to like creative writing and like the responses.

Maybe you use a specific model because you know without a lot of prompting, a lot of the guardrails, you're just going to get something out that sounds good. Like I'm just going to go, hey, can you write me like this nice script? And I'll just do it in a better way than all the other prompting where you have to be the chat-chip-tee. Hey, don't be like overly verbose. Like don't use the if that or like this, then that type of paradigm. So I don't know. I kind of see that.

Bilal Tahir (06:59)
Right. Yeah.

Right, right, right.

me.

I

can totally see, yeah, I can totally see at least in the short intermediate term. I guess where I'm a little hesitant is I'm like, what is the truly model that's AGI? For that model, why, for a model to be like basically what we think is like that at that level, it should be smart enough that if I want an Aaron Sorkin model, I should just be able to give it an Aaron Sorkin script and it should just sound like Aaron Sorkin or Martin Sokorezzi. I wouldn't need to train it from scratch.

on that. It should just be able to just take the context and just switch.

Pierson Marks (07:46)
No, and

I completely agree with that. I think where that kind of breaks down though is that you have to be explicit in your examples. And I think it's like,

Bilal Tahir (07:56)
Right.

And that's where I think there's a middle layer where most people, don't want to say, I want Aaron Sorgen. They just want something that sounds good. And maybe there's this middle layer of humans or some organization that maybe give the right context and the backstory and the prompt and just change that. maybe so it's basically the same model, but you just change some prompts under the hood or something, you know, and can have different views.

Pierson Marks (08:17)
Right. Well,

because that would be like fine tuning or like Laura's, right? As well. So, no, it's interesting. We'll see what happens. We'll see also if Gemini 3 comes out and just like, you know, blows our mind away if that comes out in November.

Bilal Tahir (08:28)
Yeah. I am excited.

I've been heart- feel like I've said this so many times because I thought it would come out like last month. It didn't. But yeah, I mean, there some rumors it's going to drop and people are saying it's a significant jump up. I hope so, you know, as a Google Bull. Also, we had GPD 5.1 come out and it was a very weird release because they didn't really...

Usually there's a lot of hype, oh my God, OpenA is gonna release a model, all that. And this one, I just found out after the fact. Like, oh wait, by the way, 5.1 came out today. I was like, what? When did that happen? So I don't know.

Pierson Marks (09:02)
It's interesting how they can manufacture hype because like, open it up, like, like how do you get the attention of all these people? You drip into like creators, like say, you leak some information, you put some hints. It's very interesting how like a few handful of companies can dominate the, everybody's talking about if they want to, like they're good at that.

Bilal Tahir (09:22)
Yes. Yeah.

Definitely. Definitely. mean, I think with OpenAI has kind of perfected the bot army. you know, I think they have, obviously they have some of their employees are huge Twitter influencers and that helps. But I also think they...

to use the kind of almost like not bots, but like people kind of incentivize them to like share and hype because they get more likes and stuff. if they've kind of like somehow either stumbled upon or actually manufactured this sphere of, you know, people talking about OpenAI. And I do think one good thing is it's gone, it's way less now than what it used to be. I remember a couple of years ago, it was just every day, every week we were just like, OpenAI is going to do this, blah, blah, blah. It was exhausting. And a lot of time it was just so unrealistic.

Pierson Marks (09:51)
Right.

Bilal Tahir (10:04)
but I think after 4.0 especially, lot of hype has kind of diluted away. People now talk about Anthropic, they talk about Google, talk about XAI, et cetera. And so I'm kind of glad that OpenAI has lost that mantle, at least partially.

Pierson Marks (10:20)
Total. Yeah.

I mean, like earlier this week too, there was not OpenAI, but there was a company that raised, I think their series B like five, five billion dollar valuation. Actually don't quote me on that number. It's a AI PowerPoint slide deck creator called Gamma. And it's that 50 employees and raised a lot of money.

Bilal Tahir (10:34)
You

Pierson Marks (10:38)
I think they're like a hundred million dollars in ARR at 50 employees, which is really impressive. and they're a creative tool. I've never really used it. I knew of it, but I don't know. Have you ever used gamma? It's a, it's a slight.

Bilal Tahir (10:44)
Right.

I think I saw the announcement.

I do think they have a lot of customers. So maybe it's like enterprise, like decks, PowerPoints and stuff.

Pierson Marks (10:58)
and

maybe globally too, but I just thought it was interesting just because it's a creative tool that I do hear some people like, love it. I went to a talk at tech week where this founder, was like pretty much the first time I learned about the company when the founder was talking about it. And there were decent amount of people in the audience that were like, I love gamma, but I tried it once last week. And it just, it's like, I don't know. It just didn't.

Bilal Tahir (11:05)
Hmm.

Pierson Marks (11:20)
Didn't feel like what I expected to feel when I was making a deck. was like, oh, I don't know.

Bilal Tahir (11:24)
Sounds about right

for most of AI products, the hype and what you can actually do.

Pierson Marks (11:27)
Yeah, it's very interesting, but they were

hyped up a lot. check it If you're listening to this and you want to use an AI tool to make a slide deck, check out Gamma and let us know what it's like. Because if you like it, maybe we'll use it. ⁓

Bilal Tahir (11:39)
More or if

you just want to skip straight to the podcast part, which is the cool part. Just use JellyPod. I mean, mean, slide deck's interesting. JellyPod, slide deck, I'm not going to give too much away, stay tuned. ⁓

Pierson Marks (11:44)
Yes, the thing that I'm wearing on my forehead.

Yes, yes, yes. Well, cool.

Okay, so let's jump into the other thing. We talked about it in the past. I want to talk about it because I'm a fanboy of world models entirely. So Marble, the world model by...

Bilal Tahir (11:56)
Yes.

Mm-hmm.

Pierson Marks (12:07)
Who's the? So Marble World Lab, definitely. Yeah. it was generally available. So it became GA, I think yesterday or today. It's with a prompt, generate a world, they go and explore. Every pixel is rendered in real time. And you can walk around in a virtual world, whatever you'd like.

Bilal Tahir (12:09)
the firefighter? Right. Was there an update or what happened?

What is their current model or is it just more at the demo stage right now? Like, how are they gonna? Yeah.

Pierson Marks (12:34)
How they going to monetize? Oh, I have no idea.

I think they just raised a lot of money, and they're just going to a research lab. And they'll figure that part out later.

Bilal Tahir (12:39)
Yeah.

Right, that makes sense.

But I wonder, feel like the utility obviously robotics environment, I mean, obviously the videos that pop on social media will be going around in the castle and stuff. But I feel like that's just more of a novelty thing and people will quickly kind of lose interest unless you integrate that to a game or a sharing thing, know, ⁓ which probably I don't think world at least this company would be focused on it. They are more serious, I think, ⁓ scientific kind of a company.

Pierson Marks (13:00)
Yeah.

Bilal Tahir (13:09)
So they're probably focusing more on the training.

Pierson Marks (13:11)
Right. I don't know. mean, there's

a lot of, I think this is a foundational layer of a lot of applications, whether that could be for physics simulations, ⁓ if you can actually do that for architecture or visualization. I could see a huge utility in being able to prompt ⁓ a world into, hey, imagine what this new building looks like from the interior, or imagine going back into time.

Bilal Tahir (13:16)
Right.

yeah, that's pretty cool.

Pierson Marks (13:36)
or in the gaming industry or at Hollywood, like storyboarding rather than like storyboarding, like you can prompt a bunch of images and walk around as like, you know, you're a set and you're the director versus building the set. But I would highly recommend checking it out. We did a demo.

Bilal Tahir (13:49)
I mean,

that's actually a pretty sick use case because I remember I was apartment hunting and doing those virtual 3D tours of the apartment are actually pretty good. But imagine you can do that at a fraction of cost, but also just the whole apartment. You can walk from the apartment to the gym, to the living area and just really get a feel for what it's like to live in that building. Or if you're building a house, you know, going through room to room. So I mean, that could be pretty sick.

Pierson Marks (13:58)
Alright.

Totally, totally. I just love this stuff because I'm just, I got into computers as a kid because of 3D modeling and like Blender and Unity and game building. And I just like see this stuff and I'm just like, wow, all I wanted to do as a kid was be able to build these cool worlds and like, I'm around. ⁓

Bilal Tahir (14:24)
Yeah. Yeah.

This is a

Speaking

of games, I don't know if you are a Steam fan, but Steam Machine came out, which is... I haven't played a lot of Steam games, but they built a whole kind of like an Xbox-like console. It's basically a PC, but just for Steam. So it's got Steam operating system and can play not just Steam games. I think it can play Cyberpunk 2077 and all the games, but it's optimized just for gaming. And if you look at the statistics, it's very interesting, Gen Z especially.

Pierson Marks (14:38)
feel busy.

Interesting.

Bilal Tahir (15:01)
up socializing for gaming because they socialize through games now mostly. you know, whether I'm not going to make a judgment on whether that's good or bad, but most kids at least, you know, they mostly talk to their friends through Discord or games now. And I wonder if this gaming will become this kind of process, like having a steam machine with poor models or something that that kind of becomes the UI where, you know, all most socializing is happening now, which is interesting.

Pierson Marks (15:27)
Right.

It's interesting because like people freak out about this. And I mean, I think there's your right to. When I was a kid, I loved coming home from middle school and playing Call of Duty with my buddies. It wasn't in replacement of, well, maybe sometimes you can consider it was in replacement of.

Bilal Tahir (15:38)
Right.

Pierson Marks (15:43)
like seeing them in person. But like I was playing sports, I was still going to school, I doing all this stuff, but I also just did this. I remember I actually had friends that I would play COD with or Minecraft with that I didn't meet in person for like multiple years. Like they were in my community, like they went to high school, but they were maybe like two years older than me. And I was in middle school and they were in high school and I kind of just didn't really know them, but we would play like zombies here and there.

Bilal Tahir (15:45)
Right.

Right, right.

Right. Right.

Pierson Marks (16:07)
But I don't know if like this is, the problem that I see is if the games are like Fortnite, Candy Crush, like highly optimized to be addictive.

Bilal Tahir (16:20)
Right. Yeah.

Pierson Marks (16:20)
and like

exploitive of children, bright colors, levels, season passes, like all that stuff, really bad stuff that's like, I actually did, it's a cliffhanger. I just need to do the next thing and just stay on versus like, play Minecraft for a little bit, which is naturally addicting, I think. And then like Portal was a great, great game. loved like Runes or...

Bilal Tahir (16:25)
Mm-hmm.

Yeah, yeah.

Pierson Marks (16:43)
Skyrim, GTA 5, I love blue games, know.

Bilal Tahir (16:44)
Right, Yeah. No, you're right. You're right.

It's interesting. mean, there's both elements to it. But I do think, I think that's a good point that a lot of older people miss. They think just because you're playing games, you're not socializing, you're losing that. But like you said, you are socializing actually, just in a different format. Now that comes with its own pros and cons, but that's another thing.

Pierson Marks (17:07)
I told my

friend, like, I don't have an Xbox or a PlayStation. I don't play games anymore. I've always enjoyed them and I get much more dopamine writing code, which is so why I turned to writing code versus playing video games. think it's, that's, ⁓ yeah, it's way more fun. like Legos. But I would play video games with my friends if I had a console.

Bilal Tahir (17:26)
It's a new game, yeah, basically.

Pierson Marks (17:33)
And I could see this in a more adult use case. Like I have friends all over the country and I think over the world, hopping on like, do you want at 5 p.m. on a Friday? You're in New York or you're in like a not a city that I'm in. But let's just like, you know, get a beer, play some COD, chat on your headphones. I don't know.

Bilal Tahir (17:38)
Yeah, right.

Right. Well, yeah, exactly.

Well, I think it's for me what's interesting is that even to this day, most of us think of

our digital lives as secondary to our real lives, quote unquote, which is our physical selves, which are going out and stuff. And then, yeah, whatever we do on the internet, that's secondary. And I actually think that we haven't quite realized, because I think the shift has already happened, and it will, or it will happen in the very near future, where our digital lives are our primary.

ways of existing and the going out to shop for groceries and stuff, that's the secondary thing. if you just, it's kind of like, you know, I guess if you imagine Wally, but except you're fit being like the North star, like if you look at a percentage of how much you spend online, it's obviously grown and I was like 40, 50, it's crazy. So, so if you just really keep that trendline going, what happens when 80, 90 % of your basically your existence is online, whether, and it's even more integrated with VR and stuff. you feel there.

It just, I don't think we are, we as a society have really understood what that's like, whether it's birth rates or whether it's a just living because, uh, doesn't matter if I'm, I want to live in San Francisco's downtown when I'm spending 95 % of my time on a computer, you know, like, you know, like small things like that. could just get a house for a 10th of a price in Nebraska or whatever. Right. And just as long as I have good internet connection, like that becomes the new thing. And so I think we'll see some very interesting, uh, social and demographic trends happening that view.

that will catch us off guard because we haven't understand this flipping of digital, our digital cells becoming our primary modes of existence.

Pierson Marks (19:21)
Right, yeah,

no, I agree. mean, think about, we're on a, we're not in the same place right now. We're filming a podcast. You talk to a lot of people via Slack, via email, even if you're in the office. I mean, if you're in an office, a lot of your work is done independently. It's done with people not in the physical space. Like when's the last time, you know, if for most people, you know, you do a deal and you get a meeting and you fly to a place or you meet up at a place.

Bilal Tahir (19:46)
Right.

Yeah.

Pierson Marks (19:47)
And

those become special moments and those are important as well. But nine times out of ten, it's just a, let's hop on a Zoom, let's hop on a Google Meet, let's a 30-minute meeting, and that's your digital identity through a computer.

Bilal Tahir (19:54)
Right, right.

No, 100%. And to be clear, mean, I'm not saying it's happening right now. I mean, there's a reason we're both in, I mean, there's something about being in a primary city that there's still alpha here. That's why, I mean, we do it. But I do think in the future that we left, that's alpha. I think one of my probably think the smartest, most imaginative creators in our generation is Balaji Srinivasan, former CTO of Coinbase. I respect him a lot. And he has this thing called the entire network state. He's been talking about this 2030.

Pierson Marks (20:07)
Right.

Bilal Tahir (20:28)
of a market. ⁓

Pierson Marks (20:28)
that's the

Singapore network pool.

Bilal Tahir (20:31)
Yeah, the Singapore.

so one of the courts, Mark Andreessen said this, and he's like, it's like the scariest thing in the world is are the words Balaji was right. And basically a lot of like smart people have said Balaji is almost always right. He's just early, like, you know, and he's done that with a bunch of things. And his big thing for a while has been the network state. And basically what this is, is like, we used to have these states, which was based on our race or we were born in the same place. And in the future, we're just going to have like almost like discord communities where

I believe certain, I have certain values about how I believe, you I should live my life, you know, and stuff. And so I'm gonna join that community and that that'll be a network based state. And so he's building this community in Singapore. It's physically located, but the primary entity is online. And so you've got people from India to Africa, to America, to Canada, to whatever, just like joining and becoming like almost like their own entity. So in the future, like you can imagine, like I can become a citizen of,

Jelly-Pot-istan, right? And it's like, we believe in creative media, expression, maybe we share certain values that are a mix of Western and Eastern doctrines and stuff, and just come up with our almost pseudo religion and be like, hey, if you believe in this social contract and you like this, country, quote unquote, join us. And if you don't, you're like, all right, I'm out now, I'm gonna go somewhere else. And I mean, that would be such a cool way to live where people can freely like...

go in and out, you know, it's like there's no borders on the internet and there's no capacity constraints. It's not like you joining suddenly our water, you know, I can't share enough water with or so there's no scarcity issues that have, you know, is a lot of why we have anti-immigrant sentiment, you know, like, you know, you're not coming here to steal my job or whatever, right? So that'd be such a cool way where you can just, everyone can just live in harmony in a way. I mean, it sounds kind of boo-boo, but like everyone just like, hey, you want a theocratic North Korean state? All right, this is you and three other

weird dudes in this discord I'm gonna make my own cool little La La Land type of country here you know so

Pierson Marks (22:28)
Totally.

You know, well, you know who's there? Remember, do you remember Travon?

Bilal Tahir (22:33)
⁓ yeah, yeah.

Pierson Marks (22:34)
So his older brother,

Otavio, he's at the network school in Singapore. He's been there for like now, I don't know, probably since the beginning of the year. And yeah, he's been there. So it's been pretty cool. I talked to him a little bit. Yeah.

Bilal Tahir (22:38)
What? That's amazing.

pretty cool.

Damn yeah I'd love to know more what life is like there. You know it's an experiment and you know it's always like there's it's an early format there's a singular figure there which is all always human nature is to make it a cult but I do think the idea is very very interesting.

Pierson Marks (23:01)
Right.

Right, Well yeah, sweet.

That's interesting. Well okay, so we talked about World Labs, talked about GPT-5.1, Gmini 3, Gamma, Waymo on the freeway. I didn't talk about that, but just so if you're not aware, if you're not living in San Francisco, Phoenix or LA.

And I think Austin and Atlanta, there's like a few other cities that have Waymos, but if you're never taken a Waymo and you're in a city that has a Waymo, definitely recommend trying it out. It's crazy, but now they're coming out in three ways. So you'll see them around.

Bilal Tahir (23:30)
Yes, I'm so excited. Go

all the way to San Jose. Interestingly, not to San Francisco airport because apparently they have their own local regulations and stuff. he the reason? Yeah, I think they have approval. They're just figuring it out. So I do think because I mean, that's one of the most requested, you know, I want to go to the airport on a Waymo thing. it's yes.

Pierson Marks (23:38)
Yeah, it's Aaron Peskin, the idiot. Sorry. He's out though. He's gone. Supervisor's gone.

Yeah, of course. Yeah. It will happen in the

next six months. San Jose Airport, you can get picked up there. Yeah, and then there was the whole KitKat kiddin' which was so funny. People were like, I was like, uh.

Bilal Tahir (23:59)
you can? nice. That's amazing. So.

my god, it's so ridiculous. And I mean, there were

a couple of local politicians trying to do like Kit Kat, click campaigns or whatever. And I'm like, you have no idea how many people are killed, let alone animals, you know, by human drivers, you know, it's just like.

Pierson Marks (24:17)
You also have

no idea that like all your constituents also use Waymo and like Waymo. Everybody likes Waymo. I have never met a person that gets into Waymo and has a negative opinion. They might have a neutral opinion like, I have like a weird aspect or things, but I've never met anybody that is like a Waymo sucks.

Bilal Tahir (24:24)
Evan.

And that's

a recent trend too. If you remember like a couple of years ago, when they first came out, people would like break your re-vemos and stuff. Vemo, cause Vemo was a symbol of, know, late stage capitalism and these tech overlords, all blah, blah, blah. There are still people who do that, but for the most part, a lot of people like now they've written Vemos, they're like, yeah, this is amazing, you know, this is technology for all of us. So why would I be anti-vemo? So.

Pierson Marks (24:57)
It's very interesting.

had a discussion with one of my friends about this and I think that like capitalism is such a great system. And one of the things that really is what it enables, it enables long bets that with when you aggregate wealth into a certain entity that no longer has to think about immediate ROI, getting like, you know, making something that in the next year is going to pay off.

Bilal Tahir (25:21)
Right.

Pierson Marks (25:21)
You can like take

on bigger bets. if you had like, let's say you split up Google into a hundred companies and each company was worth, you know, 10, 10 million, a hundred million dollars. You split them all up. No one of those companies would have ever been able to invest into Waymo and make it exist. Like you have to have enough capital to invest in building and having enough people, enough time. Cause it's been a 10 plus year project. Like the first Waymo rides were in the 2010s, like, and they were, they were good, but it was the edge cases.

Bilal Tahir (25:27)
Right.

around.

Mm-hmm.

Yeah.

Right. Right.

Pierson Marks (25:51)
It was like all those little one-off things or where it's like in this weird situation there may be a branch on the street and a person jumps out of the corner and like you know and it takes a long time and no company could really like no investor would ever put their money down unless it's a government entity which those have their own problems but like you have to kind of be able to have money and take a long bet so yeah

Bilal Tahir (26:08)
right?

Yeah, no, 100%. And I know, I mean, we're huge fans of Waymo, but I mean, apparently Tesla's new FSD version is pretty amazing too. And I know that they've started the robot taxi service in Austin, which I guess you... Oh, really? Huh.

Pierson Marks (26:24)
Right?

And no, never just go to it's here.

I have two friends that are on that use Robo taxi. There's a long wait list. One of my friends, he was on the wait list for like four months before I got off. And my other friend, they have a model Y. I don't know if it impacts if you have a Tesla or not. It seems like it doesn't. They have a safety driver, but it only is there for if it needed if it needs it.

Bilal Tahir (26:45)
But they do have a driver, right? I mean, they...

Right, yeah.

Pierson Marks (26:52)
So,

and it's cheap. He said that it's like a third to a quarter of the price of a Waymo.

Bilal Tahir (26:55)
Yeah. Right. They're

definitely subsidizing that right now. like, it's like back in the day, you could have like $4 Uber rides and stuff, you know, and then we see the VC money right now. I wish, and now look back, I'm like, I wish I'd taken more Uber rides when we were in that era. It's so cheap.

Pierson Marks (27:00)
What?

Yeah, totally.

But oh, yeah, yeah, yeah. I know. I remember I was talking to one of my friends and he was like, yeah, when I was in college, we still, taxis would line up on a Thursday night, taking us to the bars. Like we, I never experienced taxis. I just kind of always had Ubers. mean, in high school I wasn't Ubering anymore. Yeah. It's so funny.

Bilal Tahir (27:24)
⁓ yeah They rip you off. Yeah, it's gonna see I yeah

first few years. It's crazy. It's crazy how much tech has changed You know, we just take it for granted but

Pierson Marks (27:36)
Yeah,

totally. Well, mean, so there's like two, there were some things we did get. So last week we talked about LTX, some new image models, CDANs.

Bilal Tahir (27:38)
Sorry. Yes.

Yeah, it's funny. We haven't

really talked about generative media this too much this, but there's, what was it? There were some cool models that came out, was, what was it? It's funny, open routers on file now. That's interesting with the open AI. But who was, we talked about,

Do we talk about the image models that came out last time?

Pierson Marks (28:08)
Well, one of the things that I

thought, we talked a little bit about it, but one of the things that I thought was really cool that I found, I stumbled across this company called Beeple. Let me share my screen. So if you're watching this on YouTube, it's easier because we talk about, ⁓ because it's visuals.

Bilal Tahir (28:19)
Cough cough

Mmm.

Pierson Marks (28:23)
So

it's kind like video AI effects. So it kind of just accelerates the effects process via AI. So you take like a shot and then you kind of have, can relight the shots ⁓ and you can like remove backgrounds and do all this stuff. It's kind of like Premiere Pro and After Effects with AI. And it seemed pretty cool. It's like drop subject into virtual scenes and you have all that control. It's like, it's built for Hollywood.

Bilal Tahir (28:36)
Mm.

That's really cool.

Pierson Marks (28:52)
So like this guy, look, like he took it, removed the background, dropped him into a car. ⁓ you know, AI.

Bilal Tahir (28:52)
Thank

Yeah.

Yeah, these kind of image editing

flows are very cool. I wanted to check out, haven't given this model a run, but there's one called Edito, which is on fall, which is also lets you take a video and add edits to it like this, which, you know, there's different types of models with different strengths. Sometimes they're stylistically, sometimes it's placement, but the UX is super important to like, you know, I feel like, you know, that's probably very important that you can pinpoint exactly what you want to change, et cetera, in the video.

Pierson Marks (29:27)
Right. I think that's

pretty key. mean, when we talk about commoditization of models, it's going to be not the model, it's going to be interface. See, that's the key, flexible export, directive blender, Unreal Engine. So it's a combination of AI and traditional rendering tools. So you can take somebody, create them into a render, relight. It's like...

Bilal Tahir (29:35)
Yes.

Pierson Marks (29:51)
It's the pairing of traditional with AI that is very powerful for game making, filmmaking, visual effects. But yeah, if you're a creator, think, and you want to try this out, let us know.

Bilal Tahir (29:55)
Hmm.

Yeah, that's pretty cool. There's also the image equivalent of this. I don't know. There's a Quinn model that was pretty cool where you can do multiple angle. You can take an image and then if you want them to get the same angle from different shots, which is very cool, which is another way. mean, you these are, you've talked about this before, but about people who just have the talent to be, it's like, you're basically a director and some people just are better at understanding, like, you know, this is a shot I want, this should transition to this one, et cetera. And you're going to see that that creativity come to the surface.

Pierson Marks (30:15)
Right.

Bilal Tahir (30:33)
these tools.

Pierson Marks (30:33)
Totally, totally,

no, it's super interesting. And that, that Quinn image one, it's where you take one image and then it kind of creates like nine images of you and then you can like look up, left, right, down.

Bilal Tahir (30:44)
Yeah, I don't know if it does multiple or does

one, but yeah, mean, can probably do that there.

Pierson Marks (30:48)
Is that the one that

follows your mouse though? Like where it creates a 3D render?

Bilal Tahir (30:52)
⁓ I

saw that. That was cool. I think that's more of a, you make a model out of it and then it can kind of look around. But I remember I saw that too. That was very cool. That'd be a cool loading screen or something where you just.

Pierson Marks (31:02)
That is cool. Yeah.

I

think that's where the visual models, that's why I think that world models come and blur the lines between image and video. ⁓ Why would you need to generate, okay cool, like generating an image is cool, generating a video is cool, but I think instead of generating the video, you can generate the world and then you can have that world and then.

Bilal Tahir (31:13)
Right.

Pierson Marks (31:25)
Maybe you just want to zoom out your camera. You can zoom in, move the camera around. You don't have to regenerate the whole video just because you want it a closer up shot, or you just want to change your angle. It's like the world's generated it, and now you just have the ability to move the camera and pan that around the scene to make it better.

Bilal Tahir (31:27)
Yeah.

Yeah. Yeah,

is pretty sick. mean, it just matters. It's like the raw footage is the world footage. Then you're like, all right, now I want a shot from this. And you just go replay the world recording in that. Yeah.

Pierson Marks (31:51)
Yeah, move the camera around. I mean, that's how Blender

and Unity and all these tools work. It's like you create the scene, and then you have a camera viewport. And that viewport is kind of what's the projection on from the 3D world down to 2D that you rendered on the screen. So yeah, I think that will happen.

Bilal Tahir (32:09)
Yeah,

no, for sure. I mean, there's no reason why I wouldn't. And then you can get as detailed as possible with these upscale models. So I've seen a lot of cool image and video upscalers that really like, you know, make the, take that AI uncanny stuff out of the way, because you add so many details and stuff, you know, it's hard to see. I mean, it's funny, we don't really talk about it anymore because the work, we basically are, images basically are, you can't really tell if there are AI images at this point. Video, think for a lot of it, like sorenstaff, you can't,

Pierson Marks (32:23)
Hello.

Bilal Tahir (32:37)
really tell anymore, like for like, least for smaller clips. But we kind of like, whispered this critical point, it was supposed to be an Armageddon, but you know, we're just kind of like, all right, it's there now, you know, and moved on.

Pierson Marks (32:40)
Mm-hmm.

If you're listening to

this one, if you're listening to this podcast and you want to take away one thing across all 21 episodes, just remember people's memories are like goldfish and people will get mad about things. And then the next week they have no idea what they're mad about, why they're mad about it. And it just becomes reality. And so that's why technologists are amazing. When you, when you, think people, people often

Bilal Tahir (32:57)
Right.

Yep.

Pierson Marks (33:09)
say, this will never work. This will never be a thing. Nobody wants AI music, AI podcasts, AI movies. It's them projecting what they want. They hope the future is like versus like what they actually think the future could be. They're like, they're in disbelief. They're like, I don't want this to happen today. And then eventually they come around.

Bilal Tahir (33:20)
you

Autopsy.

Pierson Marks (33:31)
It's not like they don't think it will happen. They kind of are just like, I don't want this to So I'm going to say it's never going to happen. But like, I don't want to live in that world right now. And they haven't come around to the idea, but people eventually do. This is the Luddites. It's like the chasm. It's the early adopters. And then you have the, I forget the whole cat, everything, but.

Bilal Tahir (33:49)
the trough of the, yeah, yeah, no, for sure. And it's like, we don't know what we want and the stuff that actually is the gotcha, the problems like our second and third order effects, it's very hard for us to understand what's gonna happen. mean, you so I don't know. You gotta be like very non-judgmental sometimes and just let things happen and accept them.

Pierson Marks (33:52)
Yeah.

Alright.

Essentially what happens at the end of day is that most politicians know this, most leaders know this. It's that like people are going to get mad at things and they're going to be very mad. It's going to be like a day of just like the most horrible news. But next week they're going have no idea. It's going to be mad about something else. And so if you like take a bold stance on something and do something that pisses people off, everyone's going to forget. Like nobody remembers anything anymore. Like they're like, I was mad about this today. And then I was like,

tomorrow I'm mad about something else. I'm like, don't even know what I'm mad about anymore. And so everyone's angry.

Bilal Tahir (34:40)
Right. it changes. I

AI video is such a... Because when it first came out, people were like, this is going to destroy the internet. Now I see all this AI. These really funny AI videos on social media and people are like, ⁓ this is better than actual comedy. Like you have like JFK talking to like Martin Luther King or whatever, like just having skits and stuff. And it's just endless creativity, you know?

Pierson Marks (35:03)
100%. Well, we're entering into the age of creativity. Totally. Well, I think on that note, wrap up episode 21 of Creative Flux. We'll be back next week and we'll be chatting more about whatever happens. If Gemini 3 comes out next week, we'll be covering it. But maybe it's a Thanksgiving release. It's the main course. We don't know. We'll see. Cool.

Bilal Tahir (35:06)
The Cambrian explosion maybe.

Yes.

I hope so. I hope so. That'll be very interesting. All right, cool.

Pierson Marks (35:27)
Okay, well, we'll talk next week.

Bilal Tahir (35:27)
All right, take care, guys. Bye.