Doug, thanks for being here.
Yeah, thanks for having me.
How have your nerves been?
I got in yesterday, so we're still getting lay of the land. I am a little hungover, but I feel like that's part of NeurIPS. I've gone to two poster sessions. I'm really excited. The one thing I wish people did is closed labs presented more papers because it's kind of hard to see the cutting edge because it's like we're seeing some things around the edges that all the PhD students are doing, but not quite the cutting edge. But oftentimes you do see it intersect, and so that's what I wish. But I do love the posters.
Yeah. I mean, it seems like one thing I've heard from a lot of folks who have been to a bunch of NeurIPS is that over time it is less representative of the frontier just because more and more of it is moving into closed labs.
Yes. That is something that I've heard over and over and over again. And like once upon a time you literally could just walk around the poster hall and there was like, you know, information everywhere effectively. The leading edge was right in front of you. But as it's gone from academia into industry, that's natural. That happens. Like, you know, I'm going to say the semiconductor industry is like kind of a good example of this. They will only present their paper after it's like all cooked. Like, you know, they've been working on this for five years. They present something and then they've already figured it out. So yeah.
Was there an academic era of semiconductors?
I'm sure there was, but semiconductors in the very beginning was like a 1980s industry. So I wouldn't know. I wouldn't know.
Okay. So now we're in the like, oh, like Sun Microsystems, like that world of stuff, or where do you think...
Oh, you mean in AI as a representative, or...
I don't know, you were just like, oh, semiconductors, 1980s, I'm just...
So, I mean, like today semiconductors are like, if we're talking about the chip side, very mature, huge ecosystem, right? Nvidia is the biggest company in the world, right? So I don't know if we're quite in... I don't know, we're just in this era, right? And it just depends. Semis, like AI, goes deep in so many of these little verticals. So it's very hard to say. But yeah, I think we're just... what happens is if you have an insight that's too valuable to be out there for free, you just don't put it out there for free. And that's just the reality of all these open and closed labs. And it's pretty clear that if you get good enough to become an open lab, and it becomes good enough, they're probably just going to become closed. And so that's kind of this weird dichotomy. You see all these really efficient things happening, like really good optimizations, and then all of a sudden they're like, 'Wow, so good,' then they're not going to talk about their next thing. So...
Yeah. What were the poster sessions you went to? And was there anything that particularly surprised you or that you were there to look for?
You know, I really only got to see like a handful of posters, so I was just mostly checking it out. I haven't seen anything that's like really blown my brain, but I also am a tiny little AI brain, you know, my brain. So getting my brain blown out is probably... I probably don't even understand what I'm looking at. There was a really good paper I did see about like RL and SFT density that I thought was really interesting, but I'm excited to walk and see a lot more papers today.
Okay. Is there a lot of chips action at NeurIPS?
