The Exchanges

Q I know you're about to move on, but graphic designers, do they become, like, a extinct profession, or where do they go? Because you can not only create images within Playground, you can edit them.

A No, I mean, look, like, Walt Disney started as a, you know, person that drew pictures, and then he worked somewhere where he animated them, and then that just, like, evolved, and eventually got to two-D cartoon movies like Snow White and things like that. Right. And, uh, and then, and then, and then things like Pixar came up, came about, and we built like a three D rendering engine. So did all the people that, um, you know, that were people that drew the two D cartoons, they, they, they lose their jobs. And, you know, that was the end of the end of an era. Definitely not people retooled the stories that came from them were still really material to their creative process story matters more for a company like Disney or Pixar. Then the animation itself. So I think in this case, you know, the graphics matter, but I think that people will retool. The question is, are we giving people enough time to retool?

AI assessment note: “Definitely not people retooled... I think that people will retool.”

Q to make a supercharged Siri or, you know, take the search bar away from Google and then give up on all that money they're getting. Um, it's, they have some incumbency advantage, don't, uh, disadvantage, don't they? Because if they really push, push hard on AI to, you know, take up more room in the operating system, then they can crowd out some of the advantages that they have today.

A Yeah, I think Apple is in a really good position because their culture is already seemingly like one where they wait and see. And their advantages are not easily eroded because they own the hardware platform and all the network effects that are associated with that. So Apple seems like they're in a really healthy position to wait and see and build the best things, not just build, uh, kind of like aimlessly. Google feels like it's just trying to build everything, right? They're building ImageGen and Gemini and the coding and they're building an IDE and, you know, they're, and then they're like, asked all the PMs clearly to like integrate it this week into like every imaginable product. I opened Gmail, opened Docs, I opened so many different random Google things, uh, and they're, they're all trying to convince me to use AI. And I think Apple is super well positioned to just like let Google do all those experiments and then just pluck the ones that are the best ones and use its massive install base and distribution power, uh, to deliver an amazing experience, not a rushed rushed one. So I think Apple is behind, but I think that they are often okay being behind and they execute very well. Kind of from behind because they find ways to leap. Um, yeah.

AI assessment note: “their advantages are not easily eroded because they own the hardware platform”

Q I mean, that's my argument. What do you have to say?

A Well, I mean, there is that, that is, it's not too difficult to refute in part because like, just, just imagine that there's, um, video and the video represents, uh, the physics of a different world like Mars. Right. And even though there are natural physics to Mars, they don't necessarily represent the physics of earth. They represent some physics. It just happens to not be earth. And so I would say you could just pull that thread a little bit longer and just say, actually, what's really doing is it's representing The physics it understands in the videos it's being trained on, which could be incorrect physics. It's really what it understands what it's being trained on is kind of my main, the main thrust of my point. And that to a human to us, it looks like physics. It's imitating physics. And it's not, but it's not necessarily imitating correct physics. So it's really mimicking an understanding of its training data and likely, and if there's any training data, that's like cool CG or like, you know, the matrix where Neo's like bent, you know, on his back, that's not real physics of our world, but it models this training data. And I think that that's totally fine though, for a tool that's meant for creativity, that's acceptable. But can we really say that it has learned Physics. I can't say that. I don't think we can, not yet. You know, maybe lighting, but even the videos that h…

AI assessment note: “it's not too difficult to refute in part because like, just, just imagine”

Q hear like the business case here, because you know, for LLMs, that's one thing, like you can say, all right, it will read contracts, understand them, help us with back. But for images, you know, is it that it will replace design or that it will democratize design and make it available to everyone else? Like, I'm curious to hear your perspective on why that's a problem to work on.

A Right. Yeah. Yeah. You know, images are, are definitely behind in terms of overall capability and utility relative to language. You know, like, I think at the end of the day, all, all of these models at the, for the time being have their kind of narrow sort of somewhat utility, right? Um, you know, there are a lot of things posted on Twitter about how people are using language models, but the predominant use case continues to be homework. Right. Um, and then there's kind of this other one that's like coding for images. It just turns out that, uh, the predominant use case is making art and it just turns out and it's very surprising, but it just turns out that millions of people are very excited to make art and Art art can be, it's not art that you're going to necessarily always put on your wall, but it's art that could be used in marketing. Like maybe you post it on your Instagram or maybe make an icon and you use it as an icon for your app, or maybe it's a YouTube thumbnail, or it's an image that you put in a blog post. Maybe it's just a fun meme that you send a friend, but that's some of the state of images right now. It's really interesting. Imaginative art. Uh, but I think that it hasn't quite gotten to the utility of language. And I think there are a number of things that are probably coming for graphics, but I do think it's going to be about democratizing graphics for peop…

AI assessment note: “I do think it's going to be about democratizing graphics for people.”

Q of like a robot playing tennis in the style of, you know, person Y. Sure. But then you get into some really tricky questions because We have to figure out a way to either compensate these people because it's like, it really becomes like some sort of intellectual property theft. So I'm curious, like you're running a company, um, that does image generation. So how do you think about this?

A Oh yeah. Super, super interesting issue and, and really complex, you know, these days we don't really, we're not really seeing customers, you know, wholesale be like, I want it in this person's This one person's name, you know, um, which is good. I think it's a good thing. You know, Greg, Greg Rakowski is someone that I think about in this case, because a lot of people add that person's name. He makes really amazing fantasy art that on DVNR and maybe he's helped out some video game studios and stuff. I'm not sure, but, um, Uh, but his name is kind of quintessential in this debate. And, um, you know, what the reason why people, it's important to understand the reason why people are doing it. People aren't doing it because they're trying to copy Greg Kowski. They're doing it as a shortcut to get somewhere. Because if you take an image from Robert Kowski, it's not easy to articulate. In fact, there's reason why there's a phrase called, you know, picture speaks a thousand words. It's very hard to describe his art completely. We can come up with some words, but it's a, it's a vibe. It's a style. And so people are using it as a shortcut. And there's other people like HR Geiger who does like kind of more eerie, you know, worst type stuff. I've learned about a lot of artists because of how people are prompting. There's a lot. And, and, uh, you know, there's no, no easy way around this,…

AI assessment note: “users are doing this because it's a shortcut to get to a very difficult to describe style.”

Q it's already pretty good. Like it does a great job of synthesizing information and spitting stuff back. Like where does, where does it end? Where are you? Where like, because you mentioned, okay, they're working on a new model. Well, how's that new model going to be an improvement from what we have? And then where do we end up getting to if this keeps on getting better and better?

A I mean, Sam's, Sam Allman's response would be like AGI, right? Um, I think there's another, there's like another version of, of, of his belief. Let's just like one, I think one time when we were talking with Sam, um, just kind of at a dinner party, his, I think he, he said this thing that he believed that everybody would just like have a thousand employees. And I think we all thought he was crazy. Still do, by the way. Um, and he may, he may prove to be right, but, uh, I think chat bots are just like a very like sort of, Or basic primitive thing that we'll end up getting, you know, my general feeling is that one of the, I was talking to someone, uh, at open AI who was working on the robotics team there back in the day. And, uh, I was starting to get into AI robotics a little bit. I was kind of curious where things were in the general, you know, to summarize kind of like where the field seems to be, I'm not an expert, but I've talked to enough to enough people that are that, uh, you know, broadly robotics kind of asymptoted and hit a ceiling. About like three or four years ago, and the research still isn't like kind of on a trajectory. That's amazing. But the reason why I'm bringing a robotics is because I want to answer your question about where I think things are headed. Is that the, there was a belief, um, for a little while, someone at open AI at least had this belief. Can't…

AI assessment note: “the reason why I'm bringing a robotics is because I want to answer your question”