Speaker unverified: belief

Transcript context

…How did you find the podcast, and what do you feel about it, what do you want to tell people about it? Jim is Jim is super friendly, and we have to have a full episode with him at some point. But he's, yeah, I mean, he's doing amazing things at NVIDIA. I'm sure he's very happy there. I feel like, okay, Jim is this interesting mix between a researcher and a Content creator, right? So, Jim's take on Sora, I slightly disagree with, because he says it's basically a data driven world model, and a lot of people misinterpreted him, me included, basically saying like, oh, are you, are you saying that there's an underlying physics model behind Sora? And he's like, no, no, no, no, no, it's just, you know, using diffusion transformers to learn a representation of world models. It's not perfect. Then I'm like, okay, but that's a misleading analogy, I don't know. Anyway, so like yeah. So I feel this, like, pull towards, like celebrating things on Twitter, but then also trying to be realistic. Trying to present, like, what is actually the thing instead of the hype. And it's very hard to separate. And that's something that's a challenge for Lanespace. world? A little bit. It's, yeah, Sora understands a little bit of physics. The problem with this is they cannot have 80 percent physics. Like, it's 100 or 0, like, otherwise you lose confidence in the thing. So that's why you have these generated models where the chair will show up and disappear, the spoon will show up and disappear, you know, like, that's all the artifacts you see in Sora. Which is good for us for now, because we're lucky that it's not good enough yet to consistently generate all those things. At some point it will be, we just wait two years, and it will be. Very cool. Thanks for it. I love this discussion. Thanks for listening. I'm really glad to have you as a listener. Alessio and Swyx covered the Jim Fan vs. Yan LeCun world model debate in the main pod, and you can click through the show notes for more detail directly from each of them. Our third listener is RJ Honecke, who comes from a data science background, but wanted to ask about how we think about learning in public in AI, and how that informs the context with which latent space is created. Hi, I'm RJ. Shawn, nice to meet you. Nice to meet you. Do you also listen to pod, or are you just here to hang out? Yes, very much. Oh, yeah. How do you feel about it? Thanks! that. Yeah. You can, you don't need anyone's permission. That's, that's the beauty of this thing. But also no one owes you anything. No one owes you their time, their attention or, you know, or, or, or responses. And I typically try to classify these things as different modes of learning in public. Mm-Hmm. , I think I have four modes that I sketched out, but the two I remember the most are Explorer and Connector, and then there are two more advanced modes, I think like Teacher or Builder or something like that. The Explorer is where you sort of like put things out as you go along. It's learning exhaust, where you don't have expectations so that anyone will read it. It's mostly just notes for yourself. And that actually, that lack of expectations frees you. Because then you're like, oh, like two people read it. Doesn't matter, it's useful to me. It's useful to my team, it's useful to me, it's useful to whoever comes after me because I documented my work and my thinking. And that's great. And I think that's, that's the way that most people should start, which is like, just lower, you're not going to be an influencer overnight, like, it's fine, completely but get your thoughts out there, and then also, but also, like, start having feelers in different directions on what works for you, what works is a combination of what you like to And what other people want from you, and you will know when people tell you they want more from you. And so then, when you get there, when you have expertise that you have that other people don't, then you switch gears into a connector, where you are now coming from a place of authority. Like, I know how to do this right, and I will teach you, because I have done this, and I have spent more time, paid more in my dues, and here's the lessons. Thank you. And then that comes to be, that tends to become more of a polished effort that tends to become more measurable or in terms of like the impact and the influence it can get. And I think that's, that's where people start moving towards. But basically just lower expectations, make it cheap to experiment, put out a lot of stuff in different directions and see where the market pulls you. Yeah so I tend to tell developers, most developers to start on Hashnode these days. Hashnode is basically Medium if it was for developers and didn't suck. Because I hate Medium with a passion and a glowing, fiery hatred. Everyone does. It's comical how bad they are. But, I use Substack for latent space. I'm pretty happy with Substack. It's an email social network. Email is one of the most important things for people to like, come back to you frequently. So that you don't, you're not subject to an algorithm, you own your audience, you know.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.