Latent Space: The AI Engineer Podcast
00:47:00 · Shawn Wang, Jamin Ball
[00:47:00] S1: Carbon. Yeah. Carbon Argon. Yesterday? Yeah. Yes. Yes. Next yesterday. They haven't released it yet. This is my gripe at Google. They never released anything. They just announced Yes. A wait list. Yes. And then, like, goddamn it. Like, just [00:47:11] S0: I think I think it was also you saw a massive explosion of, you know, Kimi k three, the GLM series, deep seek models. You saw open weight models get really powerful and and get good quickly. So I think you kind of saw all that. But, like, it's it's crazy. Like, the far right bar, like, what was non opus five five or sonic five five, like, that has gone... That's halved or more. Maybe more, you know, in... What is that? Two, three months. [00:47:41] S1: Yeah. [00:47:41] S0: And so, like, you do... Imagine if they didn't have Opus five five or Sonnet five five. Right? Like, who knows what the revenue growth and trajectory could have been? Yeah. So, like, that's kinda why I do think it's important to be... Have a frontier class Yeah. Style model. [00:47:55] S1: And so I I will say from watching these labs for the past three, four years, they used to have more of a reserve. Yeah. Like, I will have internal models that I can just release whenever I feel like it. Mhmm. And and then I think they... Towards the GPT five, six era, everyone just released whatever they had when they had it. [00:48:15] S0: Right. Right.
00:48:15 · Jamin Ball, Shawn Wang
[00:48:15] S0: Right. Right. [00:48:16] S1: Now I feel like the reserves are building up again. I don't know why, but, like, I... It definitely seems like like, as you saw with the internal model, it it seems like they they have more wood behind the shed for stuff like this. Like, I can really... I am... I am sitting on Fable five five. I can release it the day before I IPO. Yeah. And, like Yeah. Like, that'll that'll make my IPO go well. Yeah. Know? [00:48:35] S0: I mean, look. I I I think there's I think there's good reasons, you know, that the models aren't released right away. You know, on one hand, there is just a lot of, like, safety and alignment and all that. [00:48:45] S1: But as a PM, as an as a as a executive, like, playing this game Yeah. I can play it pretty well if I have a backlog. [00:48:51] S0: I mean... And it's and it's also... Think about, you know, OpenAI's Patch the Planet initiative. Oh, yeah. Which is basically, hey. Let's use these super powerful models to go find vulnerabilities. [00:49:02] S1: Not even. It was five six. It was five six Soul. They found the PHP vulnerability. It wasn't Frontier. Yeah. Which is crazy. It was just a spam a bunch of, like like, our last gen model. [00:49:11] S0: Exactly. Exactly. And then let's let's find all these vulnerabilities like the PHP one. Let's go create the patches so that we can instantaneously release the patch by announcing the vulnerability. The challenge is attackers are using these models, using really powerful open weight models, they're finding these vulnerabilities themselves.
00:49:11 · Jamin Ball, Shawn Wang
[00:49:11] S0: Exactly. Exactly. And then let's let's find all these vulnerabilities like the PHP one. Let's go create the patches so that we can instantaneously release the patch by announcing the vulnerability. The challenge is attackers are using these models, using really powerful open weight models, they're finding these vulnerabilities themselves. [00:49:28] S1: Yeah. It's a race. [00:49:28] S0: And it's a race, and it used to be, you you had weeks, months from like vulnerability discovered to like exploit. I mean, now it's collapsed, and it's basically zero. You know, it's zero days. Like vulnerability announced or discovered, exploit available, like it's almost negative. And I think in OpenAI's announcement of the PHP thing, I think hidden in there was, I think they basically said like, oh, and we we... There are more serious vulnerabilities that we know about that we're working on patching right Yeah. Which, I mean, more more more more critical than, you know, something that touches 70% of the Internet is is is pretty serious. [00:50:08] S1: Like, the 70% of Internet is just a proxy for old. Yeah. Yeah. Yeah. Of course, it's older, and it's gonna be more vulnerable. That's right. Yeah. Of course. Yeah. You know, that's a good way phrase That's a... Yeah. I think, there's a couple of takes. I'm not that interested in, like, dwelling on on that stuff only because there's gonna be smarter people than me on that. [00:50:25] S0: Yeah. Of
00:50:25 · Jamin Ball, Shawn Wang
[00:50:25] S0: Yeah. Of [00:50:25] S1: course. The... There's one angle of this, which is kinda funny, which is like, well, we're spending a billion tokens finding all the vulnerability, and we're gonna spend a billion more to fix the vulnerabilities. What if we just didn't do either? And it... Which would go like net the same position as what we were at. Yeah. Yeah. [00:50:40] S0: Yeah. Well, I think cat's out of the bag. [00:50:41] S1: Yeah. That's out of bag. [00:50:42] S0: You can't put the toothpaste back at the tube. [00:50:43] S1: So that... But that is one argument people make, like, blame the AI instead of blame the... They they kill the messenger. Right. The the other part, which is interesting, is like, I I... And this might actually be an issue for basically nation state actors Mhmm. Who actually had all these exploits already. Yeah. And now, like, while, you know, civilians are finding them, goddamn it. Yeah. Yeah. Now I to work harder. [00:51:03] S0: Yeah. Exactly. [00:51:04] S1: Like, all my zero days are gone. Like, I I had them for ten years. Like Yes. Whatever. Yes. But anyway, so so these are fun anecdotes or consequences of this. If you want to really do super Straussian and Machiavellian, if you're a nation state actor, what do you do to to to preserve your edge? Because, like, you cannot have the channels that you're relying on lock up. Yeah. Then you then then you send a Saudi decoy Yeah. To put the thing behind your TV. That's crazy. So that people... So that get distracted by that, so that your real spy Yeah. Gets in.
00:51:04 · Shawn Wang, Jamin Ball
[00:51:04] S1: Like, all my zero days are gone. Like, I I had them for ten years. Like Yes. Whatever. Yes. But anyway, so so these are fun anecdotes or consequences of this. If you want to really do super Straussian and Machiavellian, if you're a nation state actor, what do you do to to to preserve your edge? Because, like, you cannot have the channels that you're relying on lock up. Yeah. Then you then then you send a Saudi decoy Yeah. To put the thing behind your TV. That's crazy. So that people... So that get distracted by that, so that your real spy Yeah. Gets in. [00:51:40] S0: That's just some war games. [00:51:42] S1: Yeah. I I I watch too much TV to to to think that what we are told is the full story. Yeah. Yeah. [00:51:48] S0: I think that's fair. Right? I think that's fair. Too much is at stake. [00:51:51] S1: Yeah. It's too much at stake. Okay. So coming back to the actual, again, finance, like, I really want to be a primer for everyone Yes. Who are are new to the economics of models and model labs and what what have you. One thing that have... That people have said is that the R and D per model. Let's say, like, what does what does Fable five... Sonnet five Opus five cost? Mhmm. Let's call it a billion dollars. [00:52:15] S0: Mhmm.
00:52:15 · Jamin Ball, Shawn Wang
[00:52:15] S0: Mhmm. [00:52:15] S1: That lasted me six months. Mhmm. Right? Because you just said the dep... The depreciation, it lasted last six months. Yeah. Yeah. And so isn't that worrying? Right? Like, about, like, well, these things are the fastest depreciating assets in history. Yeah. This is something that me and Saragol have have talked about before. Yeah. Yeah. Yeah. Yeah. And I think the the net is there's no, but, like, it's a qualified no Mhmm. Of, well, as long as my next model run is better than the than the than the current state of the art, we're good. Yeah. Yep. Yep. But it's a little bit of a house of cards if if the next one doesn't work out. Yeah. Which briefly affected OpenAI. Yeah. So a lot of people know by now that, you know, like, GPT four five, the the big the big run wasn't wasn't that successful. Yep. And that they, you know, they... They've they've since since fixed it. But, like, that slump Yeah. In a in a run or a failed run Yep. Can really, really screw with the the ROI on the thing. [00:53:08] S0: And look. I think there's a really good analogy to the trip industry. And I remember speaking to someone much older and much smarter than myself who did a lot of semiconductor investing in the eighties and nineties about all these chip companies. [00:53:22] S1: They've seen it all.