人工智能的自动化引擎：模型能力、系统架构与未来基础设施的探讨

Diego: Like how can AI be so unbelievably smart? How can we solve Millennium Prize problems in math, but still not automate even the most basics of work, like really basic rote stuff? We have this supercharged engine of automation that just does not have the right plugs and stuff to plug into all of this economically valuable work.

And if the whole company of TypeSafe disappears, maybe it'll take a year or two for people to truly catch up. I actually don't know how long it'll take. If model quality matters, then we are going to be in a very good position for a long time. But it's done, right? Like this has changed the path of technological history.

Host: Before we get into today's episode, I just have a small message for listeners.

Thank you. We would not be able to bring you the AI engineering, science, and entertainment content that you so clearly want if you didn't choose to also click in and tune into our content. We've been approached by sponsors on an almost daily basis. But fortunately, enough of you actually subscribe to us to keep all this sustainable without ads, and we want to keep it that way.

But I just have one favor to ask all of you. The single most powerful, completely free thing you can do is to click that subscribe button. It's the only thing I'll ever ask of you. And it means absolutely everything to me and my team that work so hard to bring this space to you each and every week. If you do it, I promise you we'll never stop working to make the show even better. Now, let's get into it.

Host: Okay, we're in the studio, a special occasion because this week Diego, my good buddy, launched Jev and it's been taking over the complete timeline. How do you feel? What's it like to be you right now?

Diego: Emotionally, never been worse. I'm a ragged corpse of a person right now because there's so much going on, and I'm a technical CEO, so I have a lot of fires to fight.

But mentally, I feel—I say this all the time, and I've been saying this kind of for years in my over-under events—like I feel like the entire AI field is like one of those carnival house of mirrors and everyone is just insane, saying the weirdest stuff that doesn't make sense. And it feels like for just this week I'm in sync with reality, like, "Oh, people see it now." AI can be so much more than what was once thought, and yes, an AI-based economic revolution is back on the table. This is awesome. I'm so jazzed the developers get it. And I want to show my internal gratitude to developers.

Host: I'm so jammed about the community and everything. It's so great.

Host: Yeah. You were saying yesterday that you decided to prioritize the town hall and not a bunch of VIP investor-type people because you wanted to make sure that they are the people that get your most attention, right: the engineers, the developers.

Diego: Yeah. It felt a little like, "Oh man, I'm talking to like really important people right now." I probably shouldn't reveal who, but it feels a little bit dirty for me. I'm perhaps overly genuine in things. It feels dirty if in my gigantic calendar events of people to talk to, the community isn't one of those. And actually, in my ideal world, it would be community all the time. I was thinking, should I host a town hall while walking to your studio? And I'm like, nah, that's too crazy.

Host: Yeah.

Host: Well, you guys have been hosting town halls on the Discord. Discord is now 100,000 people. Your Twitter—follow these stats. So, holy—

Host: Your Twitter's blown up. You know, it was really funny cuz at AIE you were like, "Follow me, please," and then you didn't provide even your handle.

Diego: I'm a noob. I'm a noob.

Host: But no, that's positive aura that like you don't know how to promote yourself.

Diego: Someone called me out when I posted like, "Holy, we're all three trending topics," and then they're like, "That's a personal feed." Of course! Of course, to you, yes, because it's what you clicked on.

Host: So congrats on everything. We'll talk about more details as you have them, but for people who are living under a rock or just want the definitive thing: What is Jev?

Diego: Let me think of a—that's a hard one.

Host: Okay. And I'm happy to reask if you want to. I'm happy to just jam on it.

Diego: I will say the first thing that I'm relieved about with this question is now I don't have to answer that question to my parents anymore cuz ChatGPT can just explain it.

So the way I see it is we need a new class of models. We're not attached to naming that class of models. The most accurate name we've come up with is System 1 models. There will be reasons, but there's a reason why we don't call them decision models because System 1 is beyond that. That's all I can say. We didn't expect this to be our big launch, so we have stuff in the tank.

Host: You should have said low-key research preview.

Diego: It kind of was, right. It kind of was. But there's a class of models that we describe them as machine-native System 1 large programmable. I think this is the class of models where the goal is for code to be the consumer.

As opposed to pre-trained large language models which are meant for autocomplete of the internet, or RLHF models like chatbot instruction-following models which are meant to reply to text, or RLVR (it's in a weird gray area with RLHF)—these are meant to have things that directly are consumed by code, hence the name TypeSafe.

So the thing we really, really, really want is to have AI be as powerful as possible, and we think the way to do that is to integrate it with software. And we are designing everything—beyond just the outside, the deep internals of the model—to be optimized for software. So number one, Jev is our first large programmable model, or System 1 model, whatever you want to call it. Jev is meant to be optimized for intelligence per dollar. Hence the name Jev, you know, Jevons—

Host: Jevons paradox.

Diego: Yeah. And it's optimized for intelligence per dollar. I love this debate with people about what is the most important between reliability, cost, calibration, and speed. And Jev will be the name of models that will be on the frontier of intelligence per dollar. There's other ways to optimize it like ML, or at least if you're good at ML it's all about tradeoffs, and we are just going all out on that.

Host: Yeah, and to me calibration is one of the new things that people weren't talking about as much. We've done an episode in the past with Clémentine Fourrier of Hugging Face where they were like, "Yeah, actually..." and this is your whole argument about RLHF: Is their mode collapsing towards what you want to hear the most or what is most likely, instead of their own internal confidence about a thing?

Diego: Can I soapbox on that for a second? I've heard that your audience is the most technical, so I actually want to get into that. And if I went through extreme precision to make sure everything in our launch video is accurate and real, apparently that's very unusual. One of the things that no one paid attention to was the downsides of RLHF, in particular, mode dropping.

Host: Mode dropping / collapse.

Diego: It's the same thing. It's the same thing. And I want to have a blog on this eventually, but I want to tell as many people this as possible because I think it's a very interesting thing. So the spicy take: I believe in Yann LeCun a lot. I think Yann LeCun's takes are actually among the closest to—

Host: What about this?

Diego: Well, should I address this now or should you go and—no, later—I actually think that among takes—

Speaker A: Yann LeCun is among the most accurate. But he has this very famous/infamous slide about "LLMs are doomed." Okay.

Speaker B: You know, like that one where he has like a pie chart with like a tiny part, tiny little thing, and says that as you increase sequence length, the probability of it making an error goes up.

Speaker A: Yes, this one. This one, I love this one because it's one of these things that seems mathematically obvious, but is obviously wrong, right? Like it's mathematically obvious, but it doesn't empirically hold. And this is my favorite thing to teach people about, like where...

Speaker B: What's the disconnect exactly?

Speaker A: And may I, or you want to tell me...

Speaker B: About mode collapse?

Speaker A: Oh no no no. So mode collapse is related to this.

Speaker B: Yeah.

Speaker A: The disconnect happens because if you are in a mode covering or a calibrated distribution, you are not overly punished about having outliers. You'd expect some amount of the time you'd be out of distribution, some amount of time you'd be in distribution. That's what happens when you cover the distribution. This was like models before GANs: they made blurry images, right?

Instead, GANs mode drop. They drop the minority classes and just do the really common ones. And this is why this effect doesn't happen, right? Instead, in order to generate really long strings without making errors, they need to be extremely conservative because it's really easy to see when an error happens. It's very hard to see when like a subtle thing that looks correct happens. And that calibration is like total poison into like the probability distributions of strings.

Speaker B: Yeah.

Speaker A: And it's a nuanced take, and I think that this is why this doesn't happen, and this is why strings are so bad at decision-making, or overloading the string models for decision-making is like a bad time.

Speaker B: And while we're on the topic of Yann, do you agree that his fix, which is like a world model like a JEPA-type embedding thing, is the right solve? Basically, one of the reasons that it could fail is because you're trying to reason over token outputs and then just looping back again and continuing going until you reach like an end of sentence. And his solve is JEPA, right, which is Joint Embedding Predictive Architecture. So is that the solve, or do you have a take on that?

Speaker A: Oh man, I probably shouldn't talk too much about the insides of ML. But I will say that my brand, other than unhinged, is practical. Even my take here is practical. Am I a scaling law fan?

Speaker B: Depends.

Speaker A: Scaling laws tell you how much better you get at a thing for amount in. A scaling law does mean exponentially more resources for normally sublinear gains, which looks to be a bad investment unless those linear gains are really, really valuable. But to me, it's all about like: what can we do with what we have to make the biggest possible difference? I can curse?

Speaker B: Yeah. Yeah, yeah. We're approved for adults.

Speaker A: Hell yeah.

Speaker B: And also we have a scaling law thing if you want to go into that later.

Speaker A: Oh, I could if we... That part is not super relevant right now. Actually, if you want to go into my bitterest lesson, I think that's more relevant. But like to me, I'm all about pragmatics, and I think that the JEPA stuff is really cool early research. I really love awesome research. Is it practical yet? Probably shouldn't say.

But there's just a lot of... I just think there's like so many diamonds in the rough left all over the research world right now that haven't been polished because people don't know how to like do the right task. And I think that what our launch did—it kickstarts us as a company, like yes; will it be great for us as a company, yes. But I think it's going to be like even greater for this direction of programmatic AI.

There was going to be a gold rush on top of us because software is supercharged, but I think there's going to be a gold rush parallel to us as well on like all the different ways we can expose things to make software more powerful so people can make even cooler stuff. And then we are back to like early internet energy, you know? And I think that's why Twitter is just like Jeff Jeff, you know, it's like...

Speaker B: It's inspiring because it's so different than what we're used to, which is: "I'm sorry you can't do this, but we do scaling laws and only the big labs can do it." Right.

Speaker A: That actually... I'm going to tangent if that's okay. I think you might enjoy this.

Speaker B: We're like five tangents in this.

Speaker A: Yeah, I get lost on all my tangents.

Speaker B: This is going to be horrible for the listeners to figure it out, but they're going to figure it out.

Speaker A: Yeah, we could edit this. So, popular thing on Discord that people keep asking me, and I haven't had the time to explain it yet, is: why am I opposed to safety alignment, and why do we not refuse?

I'm not opposed to safety as a principle, but I think that safety alignment is generally misaligned with users. And refusal is just like obviously a type error. Like if you're a human being and you're chatting with a bot or whatever, you're Claude coding and a refusal happens like "I'm sorry, I can't read DNA.py", that's an annoying time. It's annoying, right, but you can work with it, and you're forced to work with it because of Stockholm syndrome. I have stories about that too; I need another tangent deep in here.

But like if you ever want this in a dependency running in the background, what happens if that refuses? What if someone else is using that dependency? They don't know what that system is. Like you want the software to just stochastically break because a user sent like a weird message in there? Like that is straight up insanity.

It's coming from a place of like people who do not understand software, do not understand programming, and they are obsessed with—like I believe this—this horseless carriage of like AI coworker, instead of unearthing the full power of AI.

Speaker B: Fair enough. You want something that is the core kernel that is usable everywhere.

Speaker A: Yes. Exactly. Like the cognitive core, right? And you need this thing to be like so general, so optimized for its use cases. You want it to work on all the future use cases, all the weird stuff that people are doing. You know, we obviously didn't train on any of that stuff. Is it surprising that it works? No, cuz we trained on weirder stuff, my friend.

So, but one tangent up about safety alignment: safety alignment makes sense for a product in my opinion, for like ChatGPT and Claude. What makes safety and capability alignment different is: capability alignment is about doing what the user wants. That is sick for software engineers. They want their thing to do the thing, and the more predictable it is, the less they have to test and play around with it. It is not anywhere close to that yet—it could be—but there's so many more nines of reliability that we want in order to make it so good, like a database query, that you don't even have to think about it. It is just there when you need intelligence.

Safety alignment is like the opposite of instruction following. It's when you want to follow someone else's instructions, like...

Speaker B: OpenAI's.

Speaker A: Exactly. And this makes a lot of sense for a product again, like ChatGPT. If they don't want to do like some not safe for work roleplay with ChatGPT, that's on them, because like maybe that's what their users who have parents and kids want. Like that's fine, but in an API, that's nuts, right? That's completely unacceptable because people need to program around this, and that is so anti-user that it's... I can be an angry person, so I should try to calm down.

Speaker B: It's people get your passion and...

Speaker A: I think it's really good. Uh the the one pushback I'll give you is like what if we use it to kill people, right? Like that that is the actual like the not safe for work thing. It's private, personal, whatever. But like yes, like we will use it in war and like that that is something that companies can reasonably prefer their APIs not be used for.

Speaker B: I get that. I think that there's like pragmatic places where that opinion can be held. I don't think the foundation of like a general purpose technology is that place. Personally like would I prefer that our stuff is not used to kill people? Obviously. Would I prefer it's used for like all sorts of like great stuff in the world? Obviously. Will I put my thumb in the scale for that? Yes. But will I do it at the technological layer? Absolutely not.

Because that will fracture the intelligence. Every single time you need it to overfit to some weird stuff, you're fracturing its intelligence more and more. And like these things are fractured to the like they're so darn fractured right now.

So and as a furthermore thing, to me it's like I think intelligence will be more like a database than a coworker. Like I don't think it's up to databases to add checks on whether or not they're used for like what's something that's not great. You know, like CIA—actually I don't know what the CIA does really. You can imagine killing people who are not even bad or whatever.

Speaker A: Mhm.

Speaker B: And like I don't think it's the database's responsibility for that. And furthermore, like a thing that has been weird to me is when people like sign up for our thing on Slack and they're like, "Hey, we're going to deploy this. Can we deploy this thing?" I am just like my brother, we are an API, you are a developer. It's none of my business, right?

Like you shouldn't know what the whole task even is because it should be decomposed into small things. We shouldn't be able to know what the downstream users are doing and that is like a good boundary to give software engineers maximum power. Ideally, they use it for the good stuff and ideally we can like help them and like we've talked about like doing open source and charity and all of that. We have absolutely no time for anything else right now. But like they will get any of that bias out of the technological layer as long as I'm in charge.

Speaker A: Yeah, that's great. Uh while we're on the topic, let's also briefly talk about your privacy stuff, terms of terms of use, which got a little bit of misunderstanding. I just want to clarify that up front. I think it's probably takes two sentences from you about like you will not you're not being that restrictive about your API. Like clearly ideologically you take your role as a platform very seriously.

Speaker B: Yes. Uh yes. I don't know what you're referring to, but like this was I've seen a couple of things about like benchmarking. Like obviously we're not stopping people from—Oh man, I should be careful about what I say. I'm realizing—

Speaker A: No, you said you said it publicly that that was in the preview period. You didn't take it out for the launch and now you're going to take it out.

Speaker B: The team is doing stuff that I'm not even aware of. So it's great to know the team communicated that. I asked them to check in with the lawyers about that. Like we are obviously not stopping people from doing that type of thing. I'm extremely in favor. So I'm extremely anti-public benchmarks. Um I'm medium about private benchmarks that are proxies. Um you know—

Speaker A: so are you worry about uh saturation or like uh training on public benchmarks? So it's like easy to cheat.

Speaker B: Um not only is it easy to cheat there's a lot of in so uh I think that we are or anyone who's like competition with us that you know vaguely there is like you could say like—

Speaker A: we got 50 Jeff clones. Yeah.

Speaker B: Well, sure. So the let's say that there is competition or let's just say that there's let's just assume that there's an industry 2 years from now of people who are doing similar things to us. The thing that we are selling is intelligence per something per like dollar or per second. The no like people obsess about the cost and the speed. I believe that that is it's cool but like the thing that matters is the intelligence. Like the cost and the speed are like are bad things. You know, you're paying them for something and you need the thing back and the intelligence is what truly matters.

The problem with intelligence is that there's a geniqua to it, right? Like the the good model smell like the thing that happened after we launched of like 2 hours later that actually went way bigger than the video which was like holy—

Speaker A: this is actually usable.

Speaker B: Well, it's you know, you know, like beyond that, you know, like the—

Speaker A: the launch was crazy

Speaker B: and people could really sense how hard we care about that and that's truly what I think the long term of this is and I think public benchmarks are antithetical to this like they are a way to get people trust in intelligence because intelligence has a genic but the public benchmarks are extremely extremely gameable. Even if they try not to, they still will. You know, like back in the old days, every lab had a team to collect data that looks like MMLU to make it look better, which is just benchmarking, benchmarking with extra steps.

So I believe that in the long run it needs to be vibes and trust until you put it into a workflow and evaluate it for that workflow and measure it and have your own sense of like how it does on the exact workflow that matters. And our job is to keep moving the nines of reliability. This is like an everpresent part of our of what we need to be doing as a company and we need to do everything to have people know that this is something we care so much about. You know, like if we wanted to, we could have released Jev like a year and a half ago if we wanted to be dumb.

Speaker A: Oh, like like the you know my bitterest lesson, right? Like architecture and Yeah,

Speaker A: I'll bring it up since you since you talked about it uh here.

Speaker B: Hell yeah. Um like you know Sutton says that algorithms beats compute very roughly. Um data matters way more than compute obviously and doing the right task having a northstar is is is the hardest most important thing. This has happened in LLM land twice so far right maybe 2.2 times. You know there's RHF which like shifted the task to instruction following. No one realized that that was possible. RLVR did like a tiny little like edit to the to the direction and now us right RLCD we have a new task and the goal is you know programs in the loop and yeah data matters so so so unbelievably much like I can't I can't emphasize it less

Speaker A: yeah you consider yourself a data lab rather than like a model lab is that something that's the wording that you guys use

Speaker B: we are we will always like care so much about data. Um to me model capabilities means data. Data is so unbelievably complicated and that is what gets nines. Like you have no idea how how much data can shift everything. Data is so important. Yeah.

Speaker A: Holy crap.

Speaker B: So if people are looking for a job um we are hiring infinite data people actually infinite.

Speaker A: What is a good data person like? um you know clearly somebody who cares about reading through the the transcripts of uh whatever you said for example that you all your data is synthetic y but uh that's only like that's scratching the surface right like it's not like synthetic so what right synthetic but we have a people with a lot of taste and a lot of care looking at looking at these articulating what's wrong going back regenerating is is that what a good data person is these days

Speaker B: let me try to figure figure out how to like it's it's super complicated and like I literally onboard the data people with a talk that I assume is longer than this podcast will end up being. So I will try to say like the high level of it. So number one we don't do the kind of synthetic data that people kind well I'll do actually number is zero. um data and synthetic data depends on your task like the shape of your data the shape of your task changes the data like RLVR's data is kind of

Jeff: Environments, right, yes. Um, RLHF is the human feedback, you know. Um, each task has its own unique kind of data, and we of course have our own unique kind of data, right? Um, so number one, we have that. Uh, number two, the reason why we don't want to train on our users' data even if we could, right? Like we could probably ask for any terms right now, and I don't know if it would make a difference. We truly don't want that. Because no matter what, the real-world data has so much bias. There's like a power law of people asking the same things, where you'll end up overfitting to it and fracturing to it and all of that.

Jeff: And number two, we are aiming for like a complete sci-fi future years from now where these models are going to be the general infrastructure layers—layers and layers and layers, and deep down the stack to things people can't even imagine. Like I would like to think of our model kind of like UDP as LLMs and TCP as our models. All sorts of stuff can be built on top of that. And we need to be able to nail those futuristic use cases such that software developers can actually build that futuristic stuff. And the way to do that is even if we had all of the data of the present, we would just overfit to the present and then it wouldn't work.

Jeff: What we need is to—it almost feels like they're like artists, you know, they study this cognitive core. Our cognitive core is way less jagged than anyone else's. And then they find the jaggednesses and then they address them surgically in a way that—and you can never perfectly do this, right—but they do it in such a way that it addresses it in every single possible dimension...

Host: General case rather than the...

Jeff: Exact. And that requires a lot of intelligence every time.

Host: Okay, so we mentioned a little bit you sort of criticized my thinking as very RLVR-influenced, which is very fair. Um, let's actually mention RLCD. Um, which obviously you have some secret sauces. To my knowledge, you've never actually published a paper or anything like that on it.

Jeff: No. Right. No, not yet.

Host: But what should people get from this? Can you give people some confidence that you're just not making up jargon for the sake of sounding cool, right? Like one thing for me is like calibration I do think is well understood because we've covered it on the podcast, but I don't know what you mean when you say RLCD versus what people are familiar with.

Jeff: It's a great question, and actually I will give a related question: what is RLHF?

Host: Right.

Jeff: And actually RLHF means multiple different things, right? Like there's the RLHF of the original—I think it was Paul Christiano teaching a robot to backflip or something like that. Wasn't there something like that?

Host: Was that it?

Jeff: That was the original PPO paper, but I don't know.

Host: PPO is not necessarily from human feedback if I recall. Okay. But I believe it was an OpenAI alignment work that could teach hard-to-specify outputs like a backflip. I'm not 100% sure. And then there was actually Learning to Summarize. This was work by a bunch of the team that helped with Instruct and co-authored the instruction following paper, which was doing PPO on language models.

Host: This is the—sorry, I'm trying to manipulate this thing. Uh, this is 2017.

Jeff: Yeah. I'm not 100% sure, but that looks quite right. If it has a robot doing backflips or something like that, that might be it.

Host: Yes. Okay, cool. I guess I got it right. Hell yeah.

Jeff: There you go. Yeah, that's the one.

Jeff: So the idea was, can you do ill-specified things with it? So that's version one. Version two was the Learning to Summarize work that OpenAI did, which is actually PPO on language models to do something somewhat ill-specified. This is another thing that people refer to as RLHF, which I did not co-author. Oh, Dario's there. Cool.

Host: Hell yeah.

Jeff: And Radford. Yeah, shout-outs to Alec and Ryan. Love them. But the thing that I refer to RLHF is the... Oh man.

Host: I have comments on that.

Jeff: I have comments on that paper, but we're so tangents deep. Yeah. So the thing that really got to me, the thing that I'm calling RLHF is the task of instruction following. It's not about the PPO, that part doesn't matter. It's about setting a north star of this being a valuable direction. It's kind of like the Bitter Lesson north star, and for us, RLCD is this new task.

Jeff: And it is not—I don't see it as jargon. I try to communicate with precision. It's just that, hey, here's another north star, just like DPO and all of its descendants also do RLHF despite not using the algorithm in that paper.

Host: Mhm. And so clearly stating the north star is... being programmable AI is one word that I really catch on to, removing the human in the loop, because RLHF is tuning for this so that you can automate everything.

Jeff: Yes.

Host: Did I miss anything else in the thesis of what the north star is?

Jeff: That is right. I am overly nuanced in my communication. The one nuance is that we need to be practical. We need to be aware of what language models can do really well, like what AI can do, right? Like there could be programmatic types that are sick AF, but if the technology is not ready for it, it sounds totally arrogant.

Host: No, no, no. I strongly believe you.

Jeff: Cool. It sounds arrogant, but I felt this way since long before I even had a company.

Host: I can vouch for that.

Jeff: I've been talking about it for like 3 years.

Jeff: Yeah, I've been talking about this for so long, and I've been saying it because I thought it would have been easier. Um, they say they do not do things because they're easy, they do them because they thought it was easy. So something like that. I thought this whole project would take a week. And I was unbelievably wrong. So I am so sorry to everyone at OpenAI that I thought, "Man, I'm solving this right now."

Jeff: But like I think that the tragic thing is when—well, I think overpromise underdeliver is tragic too, and AI is super extreme on that axis, and I think RLVR is like the main—well, both RLVR and RLHF are extreme perpetrators of this. But like to me it's like there's just so much potential there. AI is clearly so smart.

Jeff: I love this in my talks, you know, when I ask people: how can AI be so unbelievably smart? How can we solve Millennium Prize problems in math, but still not automate even the most basics of works? Like really basic rote stuff that doesn't take extremely smart people to do. It's not a satisfying job. There's other things these people could be doing, but yet we need them to do this super basic, unsatisfying stuff because we can't automate it yet.

Jeff: But we have this supercharged engine of automation that just does not have the right plugs and stuff to plug into all of this economically valuable work. And you know, like if the whole company of TypeSafe disappears, maybe it'll take a year or two for people to truly catch up. I actually don't know how long it'll take. If model quality matters, then we are going to be in a very good position for a long time, but it's done, right? Like this has changed the path of technological history. And we will be exploring that space as a field.

Host: Yeah, I definitely agree with that. You've created possibilities. So I think if I can paraphrase so that people can also understand: you should not take the success of TypeSafe and Jeff as just like, "Well, you know, that is a new model type, now we're done, we go back to business." Like no, actually there are like five other model types that you should be exploring, and let a thousand...

Speaker B: ...flowers bloom.

Speaker A: Absolutely. Like, and some of which you will probably also... early internet energy. I think it's back to tech utopia. You know, it's no longer like, "Oh man, sometimes my coding agents work, but all of the best ones are hoarded internally," right? It's like creation is back on the menu. You know, though, it's going to be a wild ass world. And, you know, buckle up. I'm so, so jazzed about that.

Speaker B: I mean, and now you have the funding and the momentum to do whatever you envision there, which I think is very gratifying to see you have after, you know, so long of saying these things, but like actually show the world.

Speaker A: I know. It was such an interesting thing to be a tease the whole time. Like my talk felt like it was a cliffhanger cuz I didn't say how the automation would occur. Yeah.

Speaker B: Sean reviewed our manifesto and he's like, "It's a little bit vague in these parts, and you know, like what's step one? What is the intelligence?"

Speaker A: I asked you for model and you were like, "Yeah, model coming." And well, I mainly objected to the word composable, but Bill Pron got is fantastic. Thank you. We've really rallied around that. I'd like to think we're not entirely a cult like some companies are, but we are like jazzed about what Butter doing and my brand is being practical and we are all like so super duper practical. It's really great.

Speaker B: Yeah. So here and by the way, here's the secret master plan, right? The shape of machine native composable AI. It...

Speaker A: ...was your idea to make a secret master plan.

Speaker B: It's an Elon thing. When he started Tesla, he was like, "Here's what we'll do."

Speaker A: Yeah. I'm giving official credit to you.

Speaker B: Thank you. Thank you. Thank you. But you know, you should have told me you're also going to do this model launch, cuz you told me half of the story and then the other half you didn't have the Doom demo at the time. You didn't have any numbers to give me. I was like...

Speaker A: Well, the problem is I don't believe in benchmaxing. Exactly. Right. So like it is a thing that you need to feel, and like I think that this is the way to build long-term trust even though it hurt us a lot. You know, like last year when we did fund raise, no one believed us. They wanted just benchmarks and stuff, and we're like, "We're not going to do that, we are principled, we're going to stand by our guns. That rewards bad actors, I don't give a... you know, what do you want? This is who we are and we are standing by that, so sorry."

Speaker B: Yeah. Well, in some ways I think choosing the hard path, but you end up making the company that you want to work in.

Speaker A: Right. Otherwise, if you sell out, then you're just working in like OpenAI, but with my people, right? Which is...

Speaker B: Yeah. Yeah. I mean, I don't have too many regrets on that, obviously. Like it worked out so unbelievably well. And you know, I was emotional last night when I was talking about the reasons I left OpenAI. And because it actually had to change my wording after the launch. My phrasing was: if an AI winter did happen and I did not do every possible thing I could to avert that, I would see myself as personally responsible both for the RLHF direction, which I think really widened overpromise versus underdeliver, and also not going all in on this, because I think this is where value is going to just be like printed. So, and it was really cool because I feel like the AI winter I'm worrying about is averted. You know, like AI will be useful. It'll be used for automation. It's been less than a week and like the numbers are already undeniable that it's like being used for real work, and it's the wild west. Yeah.

Speaker A: Yeah. Can you share... just if you have top of your head, what numbers are you seeing? Like what's like signups, whatever you can share? I'm actually not super on top of everything. Like the team is the ones who are telling me all of these things.

Speaker B: I'm sure it's like changing every day, right?

Speaker A: It's kind of nuts.

Speaker B: If there's a milestone that you're like, "Well, yep, that's one thing we're hoping for. We reached it."

Speaker A: I will say a milestone that we've passed is tokens per day.

Speaker B: Um, and this is not like fleeting tokens per day. This is like even at night like it's constantly churning. So, you know, machines are calling it and not just people trying things out. So that is that is so cool. A trillion tokens a day is a lot. Um so surpassing that is awesome. Signups to me don't really matter. And actually this was like a bit of a mistake we made if I'm like totally honest. People on Twitter were calling us like marketing geniuses and all of that. And um that was just us. We don't have a marketer also hiring. Um and we were just being our genuine goofy like irreverent selves. And we were we were just like offboarding people off the wait list so hard. Um our platform team is so unbelievably cracked. I think we have more up nines of uptime than anthropic while having the most unprecedented launch ever. Like that is kind of nuts. So like props to them.

Um and uh the thing we didn't realize so number one weight lists weightless signups don't matter for like a developer platform in my opinion. you know, uh I would guess that a large number of them are not even developers. So they go in, they try some queries and a lot of people don't get it because they are not programming, right? Like they're just like, what? This is not a chatbot where where's my chat GPT2, right? But if like I I haven't exactly calculated this. My sense is that if every single human being in the world like just wrote a couple of queries, that would be a rounding error compared to like one power users for loop that is just like creating value.

And the thing we didn't realize with a weight list is like we just off offboard anyone off the weight list. It doesn't matter. The scary part is rate limits. And then once people start getting value from that, uh then they just want tons and tons of rate limits because this is what software is, right? Like you spend effort up front to specify your roach task and then this wrote task creates more value than it takes to put in and then now that you have exactly yeah you run it in the background you make it a dependency to like other things you can make like higher level stuff and uh like you just create so much value in the world you know early internet people probably did not imagine like the wonder of early 2000s internet which is still not early internet but like it's it's through no offense composability um that all of the crazy stuff happens. I just really wanted to emphasize that in our manifesto. We are going for emergence. We are going for like being the catalyst. We're wanting to empower people and we are going to do whatever we can for that. Be it like Discords in our town hall with me wearing a garbage bag or not.

Speaker A: Um and uh and podcasts and and you know getting getting like cuz I want the long form, right? It is like yes, we'll get past some of the superficial things and then we'll go deep and people will really trust and understand your mission and like you know the the people that uh will resonate that will end up joining you or or you know uh uh buying you uh no sorry as as a as a customer as a customer.

Speaker B: Yeah. Yeah, that was funny. I'm sorry.

Speaker A: Sorry. I didn't I didn't mean to say that. Um but no any one one version one very flattering version of this like 36 million views of your launch video.

Speaker B: Cool. Up to 38 now. Uh yeah, I'm rounding error. Uh you know, uh Navia Stokes got 74, Fable 5 got 57. So like as far as uh and I I didn't I didn't do the stats for like original ChachiBT like which there was no video. Yep.

Speaker A: So like up there, right? Like as as far as as far as like if you were to launch a Neolab in 2026, I think you're like number one right now, which is like pretty crazy.

Speaker B: Yeah. Well, I...

Speaker A: actually would rather I do have the shirt like your favorite Neol favorite Neolab's favorite Neolab. Um I don't give a about being a Neolab. I think being a Neolab actually we have a lot of like swag that's being a parody of a Neolab one of them one of them I have is like Neolab with product which actually is not a neolab like I don't care about that really um what I care about is being a reliable dev platform so uh appreciate the comparison but like hopefully we transcend past them and we go back into like a thing you know like you know a revolutionary moment for developers and like this stable thing that people can rely on and trust.

Speaker B: Yes. Uh I mean to that end um I mean I I think that's one thing that really impressed me about you guys is is that yes you you you do talk about reliability. I thought it was mostly about calibration which like we you know we talk about RLCD uh but actually it's also about just like uptime and and uh scalability and all those things right they all sort of

Speaker A: nines it's like like

Speaker B: which time is in my

Speaker A: but that's part of it but like there's reliability in um like how intelligent the thing is like how consistently does it do the thing that you want and I think that like the the big reasoning models are very smart in my opinion they still lack reliability I think there's many use cases where you they look like they should be smart enough to automate their work. There is economic incentive to automate that work yet still they're not reliable enough as at an intern because they're optimized for different things. And so like I think that there's the reliability of being able to like trust the outputs and also we are like like there are dimensions of reliability that we are not yet at that I'm like so excited by you know like I want to automate the easy work before the hard work you know like I think that that's just a common sense thing to do. Um but to me we will be sufficient I don't know if there's a such thing as sufficiently reliable but I want to get so good that people don't even need to try the model to know that it'll work. It's like that's like what flow state is in programming, right? Like I'm just like writing queries because I need intelligence in here and you know like when for non-trivial branching I can just write it in in like a like like a type safe system one query and then get the results out and it just branches accurately like that would be so so good. Like that's that is the dream and that is like going to be like a long long slog.

Speaker B: Yeah. We're going to go into your API design in a little bit just just to give people examples and like maybe path not taken that kind of stuff. Um, one thing up the front that I do wonder about in terms of reliability is I noticed that there's no seed uh there's no uh and and so basically same input do I always get the same output?

Speaker B: If not why not?

Speaker A: Oh great question. So this is actually like a common question we have between um so reliability is actually a catch-all like whenever AI can't automate something it's due to some form of reliability could be like type safety it could be determinism it just could be like it's it's jagged right so um reliability is a catch all I just think that it's also a catch all for like what the north star is um determinism is like same inputs same outputs I do believe that this is slightly interesting for unit tests, but I believe that to be the wrong north star. I I believe robustness is what people I don't want to tell people what they really want cuz that would be a little arrogant of me. I believe that that is like the more important property. Um uh you want um given similar inputs get similar outputs. And it's kind of wild how unreliable LM are. Like a way that we test this is you put like uyu ids in you like little I think they're called nonses in in the prompt and what you want is similar outputs from all of those cuz it's truly semantically the same question and that is the part where you really want like like like that robustness is where like people get like burnt with AI making decisions. So I think that is the a super duper important property. We could also have determinism that is that is a thing that can be available as far as I can like mentally model for programmers like it it could be valuable for some use cases. So like please educate me um in comments or you but um my in in general it's easy the determinism is something you can like trade off for better cost you know like we are we are constantly wanting to be on the intelligence per dollar frontier. We are doing like absolutely disgusting things to be there, you know, like this is um I I shouldn't say this, but no one's here to stop me. Um

Speaker B: you know, if you you sign off on your own PR,

Speaker A: that is not how it works at this company. Um I believe for this week, um my chief of staff, Kay, is the most powerful person in tech.

Speaker B: And shout out to K for organizing this.

Speaker A: Holy holy She is so competent and powerful. She's incredible. Um uh I mean she sucks. Don't poach her. Um but um so I tried to be a bit more filtered, but like people are telling me don't call it a Frankenstein's monster of models, but because that has like negative implications. I think Frankenstein's monster was like the good guy in this whole I mean it was innocent, right? I didn't read it. Okay, I I'll confess. Okay, that that what one facial expression I don't

Speaker B: decent Jacob Vorti movie if you want to see

Speaker A: I

Speaker B: the adaptation anyway

Speaker A: you have no idea how little time I have right now my priorities are sleep you know

Speaker B: developers developers developers

Speaker A: developers yes developers developers developers um but um yes we we do like absolutely disgusting things to be on the parto curve of intelligence per dollar and we are going to keep doing that we're gonna be doing crazy ass stuff. And I think people really need to think outside of the box like like part of the reason why surprising is like people are thought inside the box and we continue to do that. Um as of right now we are obviously the best at this and we want to continue being the best at that whole thing.

Speaker B: Yeah.

Speaker A: So um wait where did where did we tangent from?

Speaker B: So so I asked you about will you have seeds in determinism and then you basically define reliability and like how you see it. Yes. But like determine like

Speaker A: I have an robustness example that's that's real quick. I can show you.

Speaker B: I love that.

Speaker A: I'll just say one thing. We can make a deterministic model like like we're h if people can convince us that that is a valuable thing to do and we don't have a gigantic GPU shortage. Um we can happily make all of these models we live to please. Um and re and and revolt revolute uh

Speaker B: uh

Speaker B: you'll throw over everything except you do it in a nice way and

Speaker A: so so like determinism could be on the card. just gets you less intelligence per dollar.

Speaker B: Yeah. Uh well, just having seen the trajectory of OpenAI and topic uh you will just trust me now that you will be peer pressured into doing it. Uh so like just people will want it. Even if they if you tell them they don't need it, they'll still want it. So like yeah, that that's the TLDDR of

Speaker A: Okay. Okay. I will love to maybe one day we will see how that happens. I've been told I'm um

Speaker A: they they they say that part of our brand is being unshakable and they say that that's just the nice way of saying stubborn. Stubborn. Yeah. Exactly. And I'm a very stubborn person. I don't think we could have done that.

Speaker B: Yeah. No, but so like I Okay, but I tr I like have argued with you before.

Speaker A: Yeah.

Speaker B: And you've been right about developers every time. Okay. I give up. You win. You win. I I'm sold that I've argued with you before.

Speaker B: No, no. I'm just saying like I I think that you can hold your ground while also like if I give you the right evidence, you can uh not you can sort like throw away your priors and be like, "Yep, like that actually makes sense to me." And so like, you know, just just trust your own gut on this.

Speaker A: I'll bring I suspect though that we will be GPU constrained for a very very long time. And anything that has less intelligence per dollar

Speaker A: means it consumes more GPUs for the same intelligence which is you know like our goal is not to onboard companies like it's valuable but like our goal is to have people like experiment and do weird and we need like we need to like get it to as many hands as possible and like starting like the California gold rush for that.

Speaker B: I think there is right now. Yeah. Um just a word of caution. I mean I I'll just say it because somebody is thinking about it right now which is when you say things like uh we will not commit to deterministic models. We we will we'll do whatever it takes for intelligence per dollar and we are we are facing GPU constraint. people are thinking you may quantize your models, right? Like like whatever you had at launch, you may quantize down in to reduce the quality um in order to to free up uh memory or bandwidth or whatever, right? And so uh you should probably uh have some kind of promise which you don't have to make now about like we will uphold model quality at launch people like when people when when we I mean you were at OpenAI when you launched all these all these APIs and even claude as well like uh when they first launched the models the model strings um did not stay the same model at all times right you have versioning in your models that's great but like you should you should publicly commit to some kind of like once it thing is launched we don't change it

Speaker A: we will not change our models when we deploy them. That is insane. We care about developers like like it makes sense if you're so doing something like that. Again, this is the problem with a for first party product and an API. It makes you can do whatever you want in a first party product, right? Like like more power to them. Whatever gets that experience, that is fine. With an API, you obviously can't do that. But I will say that we uh plan to move a lot faster than many people are used to model providers uh doing things. So we will be launching new models a lot faster than people think and we are not promising long-term support for the models because we think that there's lots of improvements to have. So um there is a world that we might temporarily LTS what is right now Jev 1.13.0. We might do that because so many people are using it and I know developers hate breaking dependencies. The alternative is fracturing our fleet and that is a very bad vibe for everyone.

Speaker B: Have like 100 different versions of the model.

Speaker A: Exactly. And if we're iterating very fast, there would be a lot of those versions as well. So um we we do want to have not just a LTS supported thing eventually long-term support. Um we want a really sick way of doing that. we have like research stuff cooking in that direction and I think it's going to be the most prodeveloper thing ever. Um, but it is not yet our current models and I'm not promising that we will be able to keep the exact same models. They will get smarter every time for sure. And my sense is that even our model iterations where it already is smart um it between model versions the the changes tend to be even smaller than the string models calling them twice. But when when it we go from like you know jagged to like wow um that is where the big deltas are.

Speaker B: Yeah. Uh one one thing one thing that's beautiful about LTS models is that actually you can also port them to other silicon. I don't know if you've thought about this.

Speaker A: No comment.

Speaker B: Okay.

Speaker A: So I care about intelligence per dollar.

Speaker B: Yes. Right. Um but speed.

Speaker A: What

Speaker B: speed as well?

Speaker A: We'll see.

Speaker B: Yeah.

Speaker A: We'll see.

Speaker B: I mean this is a whole part of the inference tech tree that is like I mean exploding in the past year, right? that you can move to a cerebrum and etched or whatever and get 100,000 time speed up.

Speaker A: Yeah. Like I I think that intelligence per second is like a different metric and we've even talked about like things like intelligence per dollar time second and like metrics like this. My guess on like Jeban's paradox occurring or at least the Jeb series of models. And the thing I like hunt people down about internally is like I don't care how much smarter it is. It needs to be in the pre-frontier. So like that is what the brand of Jev is. It is the best thing at intelligence per dollar for intelligence per second. We'll see. I I think that it's an intriguing thing. I know that there's many industries that are like extremely dependent on real-time stuff and they will like like intelligence per second means tons of dollars for them, but we'll see. I I I I would love to like do both and like have the market correct me either which way, you know, like like I I I I would love to be informed by people.

Speaker B: Yeah, totally. And it's not it's not just real about real time, right? It's also about scale because um at at scale every microcond is just multiplied by billions and trillions of

Speaker A: it depends on how background it's running right like if it's like a big background like database map produce query the latency might not matter so much as like the cost to get intelligence from it but like if it actually is um something more realtime like userf facing you have budgets like between 100 milliseconds and 1 millond that are like totally magical and actually even if you were below 100 milliseconds if you half that time that means you can get double the intelligence or sequential intelligence calls to have like a like a phenomenal experience. So right that is definitely happening right now. It is super duper cool. I love the intelligence per second use cases but I don't think that that will be Jeb's niche.

Speaker B: Okay. Yeah. Fair enough. Okay. When think about promise of faster and cheaper uh um typically the other uh the the trade-offs that other models are offering is faster but more expensive. Yep. Right. And so you're like one of the reasons I was thinking about why is Jeff resonating so much is that you've done the faster but cheaper side of the quadrant. Yeah. Which is very very unoccupied while holding intelligence like somewhat constant.

Speaker A: Uh yes. Yeah. Well I that that's a very loadbearing statement while holding intelligence constant. That's the hard part, right? Like

Speaker B: which unfortunately like so basically you you refuse to have to like like do any public benchmarks or you don't like any public benchmarks about it

Speaker A: and I will need some internal sense.

Speaker B: Say it again.

Speaker A: You need some internal sense of this.

Speaker B: Oh, of course.

Speaker A: Of course. We have we have our own internal evals for sure. But it takes a lot of discipline not to game those and it needs to be like a top level priority to not game them. Of course we do that, right? Like how else can we make the guarantee that our models are in the prede intelligence per dollar, right? right? Like that we're not flying blind in there, right? If we're doing like completely weird things with different costs or whatever else, you know, like how do we compare them? We plot them and get we try to figure out like what is the best for the users. So, we for sure measure them. I'm not anti-measuring. Um but it it it's extremely dangerous when you have like any alternative incentive. And this is the one thing that like I kind of rule with an iron well maybe my co-workers might think I rule many things with an iron fist but to me like not ourselves uh um about how smart our model is is one of the most important things there like we need to be truth seeeking.

Speaker B: Yeah. Yeah. Agreed. Agreed. Um okay I wanted to go over some uh details on the uh API choices mostly because this is the only podcast that will ask you these these kind of questions. Um so you have three primitives. Um choice score no. First of all no where is that from is this like a term in the in the literature or what?

Speaker A: Uh now now it is um we debated this a lot. We debated this a lot. Um it is you know it is boolish right like true false. Um it is

Speaker B: but it's continuous.

Speaker A: Yes. Exactly. Oh so so first the origin of the name is Bernoli.

Speaker B: Ah

Speaker A: yes so that's why it's even spelled that weird way that is like a subset of the name Bernoli from like a Bernoli probability right which is actually what that is. Um so that is the origin of it. We were debating this a lot. We liked

Guest: ...we liked pool. We were wanting to call it like a pool party, but then no one let me. We had a bunch of other arguments about that, and new we figured was like the best thing. Our rationale—and this is actually the same thing with Jev too—is that we think that we are like an irreverent, insane bunch, and programmers don't care. If Jev is just going to be a string, we didn't expect it to catch on or even have puns or anything like that, right? Actually, there was a lot of hate on the name internally. They've all apologized except for one person.

Host: Still holding strong.

Guest: Yes. Uh, our mutual friend. Yes, yes, yes.

Host: I respect her for that.

Guest: Yeah, yeah. She wanted Jeb to be called Meow.

Host: She would, of course.

Guest: Yes, of course.

Host: Okay. You win there. You win there.

Guest: But the rationale is we had to make a new concept for this thing, cuz if it was a bool it would be confusing to people. So actually all three of these are actually new concepts. These are not types that exist in programming, and that was intentional because they map very closely to types, but they're not quite that. A score is not an int. So if you had like Instructor or Pydantic or whatever map ints or floats into scores, you'd get a little bit cooked. And we were really erring on the side of clarity over the side of making people easily understand what's going on.

Host: I mean, don't you worry about that? Don't you want things to integrate directly into things that people are already using?

Guest: Yes. Yes, we do. And actually I think that...

Host: You have integrations with other SDKs and stuff, but you have—sorry, you have your own SDKs. But typically, for example, as a developer relations person, I would be very obsessed with like: yes, here is how you use Jev with Instructor, here's how you do that kind of stuff.

Guest: We might have that somewhere. I'm so behind on everything.

Host: Someone will do it for you in the community. Not that you're successful, people will be like, oh that's cool...

Guest: But I don't see that as binary either. I actually see success as a score, and there's always more to climb in how much we can be there for our community, just to be clear. And this section is stressful cuz I didn't review the docs and they're constantly changing, but to me scores do exist. So scores are similar to like LM judging, right? So if you want to call it like a judgment, I guess you could, but that is the way people already use this type of thing, right? Like maybe a new could be like a probability, but everything for us is a probability. And a choice is actually closest to a function call, but a function call is like an extremely disgusting thing that—if you want OpenAI juice sauce tea, we should go back into that later—like a choice is just like the right way of exposing like a switch match statement within.

Host: So it maps cleanly to an enum.

Guest: Yep.

Host: And you can choose to hydrate it into a function if you want.

Guest: Yes. And like the enum choice is the important part of that. And actually I think these map all to programming primitives, where choice maps into a switch statement on an enum, nules map to if statements, and scores map to sorting or thresholding at a greater than or less than. Okay. And this has been always what the vision is: there will be more types, and they will map into programming primitives.

Host: Yeah. Any other nuance you want to go through for literally the Jev people who are deciding to really invest in Jev? You are the expert, right? I'm just wanting to provide more background for them on API choices, how they should use some of these things like legends, confidence, how critical in your testing, just any sort of pro tips that you want to offer people when you're down at this level.

Guest: Thank you. I love this. No, this is why we're here.

Host: Hell yeah.

Guest: I didn't expect this, and actually no one has asked me this in probably months, when I was onboarding our devrel. Okay. So our model is designed for being deep in the insides of computer programs in the future. We unironically believe that this will be much more massive than anything people are even considering today. And our model might not be ready for that, but we are continuously working for that future. It will never be good enough at these shallow tasks. Sorry, we're not just going to keep on climbing the shallow tasks. We want to be deep in the guts of programs, because that's how you make software powerful. All the types inside of our—this is actually an output—but all the parts of the input, like the state, the instructions, the criteria, all of them can be structured JSON objects. That way programs can insert them in the right spot and you don't need to put things into templates.

Guest: Exactly. So I think people don't read into this part enough and they think it's all strings, and that's fine. But these are all meant—like I would say that if you're using a template, turning it into like a system message or something, you are thinking in the old way. We should be making things as easy for computers to understand, because that structure is truly there, right? Like it would be weird in a programming language to have all of your numbers in, and then you turn it into a string. Normally you do that for printing when you have a human in the loop, right? But within the computer, you want to be passing nested structure that is semantic all around. And we are really going to be optimizing our model. The model's pretty optimized for this. But the thing is, every different nested level of structure is harder to reason about. And we are really cooking hard in that direction. I think people should keep cooking that direction because it makes the code so much more legible and beautiful, and agnostic to the implementation details. It's like: here is my state, here's my function state. Think of it as like an AI function. Which subsets of my state, which is like all the variables you have available, should I pass in here? System messages are like disgusting global variables where you just put everything in there and you put all the instructions at once, and then you hope that every single instruction gets nailed instead of asking the questions in parallel.

Host: Okay.

Guest: And also I would recommend—and I truly say this not from a "makes me money" perspective—I truly recommend asking lots and lots of questions, break them down, make them smaller, and really decompose, no matter if the models can do it today or not. I believe that the biggest saving grace of what's happening this week will be people's AI code bases are going to be so much better. If you decompose problems into simple decisions, every single one of these things is extremely evaluable. An AI beforehand is a big system message, and then maybe you have another big AI to see if it actually does this. That's nuts, it's kind of crazy. That was like Stockholm syndrome, right? But that's kind of crazy. If you want to say, "hey, don't read this subdirectory" or "don't pass any API keys to DeepSeek or whatever else", that should be programmatically basically guaranteed. And you'll never have guarantees of any machine learning model, but by breaking it down, you can actually measure it, right? You can verify that it was actually called...

Host: ...and like our model, the interface itself is so verifiable, this should be like a sigh of relief, you know, it's just going to...

Speaker A: ...lead to way better engineering.

Speaker B: Yep, I think I get that. And so, you know, one of the reasons people didn't use to do this in the past was because they would just call a small LLM, right? And it's still too slow, it's still too expensive versus chunking everything. There I've done exactly this myself, right? Like I benchmarked: here's a pipeline that fills everything in system prompt and it just gets one big output, versus break it down into 100 different things. It was slower, more expensive, not as good.

Speaker A: Yep. Yep. Yep. And that happens. Yeah. It's like super inconvenient. It's unwieldy. Why not just put it all together? You kind of end up repeating some stuff between questions. So, it's like maybe like, you know, inefficient or something like that. But then it results in something that is very hard to rely on. And software...

Speaker B: ...doesn't need to run in the background. It would break my heart if our stuff couldn't run in the background.

Speaker B: Is there a way to break things down that you guys have found that works versus what you thought worked and doesn't work?

Speaker A: Interesting.

Speaker B: Because like people are just going to be exploring this, you know, now that you've said it. Like they would use this as a reference and be like, "Okay, like that's how I'm supposed to use Jeff."

Speaker A: Yep.

Speaker B: Then the question is, how do you break things down?

Speaker A: Interesting. I like to break things down into its smallest semantic unit. Like what is the lowest level thing? I try to never have... I've probably queried the model the most among anyone, and like I try to number one in my queries—this is a lot more like the way I prompt things—like I make it really really structured and explicit. And in the questions, I always... I like the back ticks, but like it works for all of them, you know, like be really clear what I'm referring to, because we want the model to be really literal, because when you program, you want things that instruction follow really really well. That is what the art of programming is, and what AI does is expanding the kinds of instructions that can be followed.

So I'm a fan of doing that. Sometimes I'm a little lazy and I have like more hybrid things, but I think that for like really big production things, you just want to like keep on adding more questions. You want to make it really easy to add more questions, you know, be really really precise about all of that breakdown, and then have the code to have the exact behavior you want.

If I could give like a tiny little example of this, is like refusals, right? Like I'm not going to talk about why we don't refuse. I might have done that already. It's like all blur. But like for refusals, I don't think you should ask "should I refuse here", you know? That's a really... I think the answer will be pretty good because like that's a system one compatible task. But I think you're way better off like asking many different independent questions about like the different situations you can refuse about, because instead of having to like just guess, you know, you can actually specify what you want.

And beautifully, you know, and I think this is like truly really beautiful: if you find a situation where it's like, "Oh, it didn't refuse because of this reason. I didn't specify this part of the task." That is awesome. That's what software engineering is about. Like you fix the bug by adding that question in, adding the threshold, maybe remembering that as a test case, and now it is just solved forever. Like your software can't forget about that like in the prompt because of context rot. It is just there, and you can like just keep measuring that forever, you know.

And if the models are not perfect at some of these things, you can choose what threshold you want for all of these factors based on real examples. It's like ML without the ML. And you can just do it for anything. And like there might be some things the model's not good enough yet, right? Like I'm a little bit afraid when I see people doing trading with the models, like automated trading.

Speaker A: It looks cool. I just think that people should leave it to the professionals. And like that's just a very hard high-level task that maybe the models aren't good enough yet to figure out. Even if they were, then it suddenly wouldn't be because of efficient market. But like that's one of those things where you can like break it down into things and just evaluate them, and you might be like, "It's not smart enough at this. Maybe we don't deploy it yet for this version, or we make a trade-off, or we err on the side of safety, or like hey, the models are not good enough at, you know, like detecting like this weird combination of like sarcasm with a VIP customer, that this is when we escalate to a human." And that's what confidence estimates are about too.

Speaker B: Okay. Great, great answer. I think one thing I'll mention very quickly, which I don't expect that you have too long of an answer for, is well, no, no, no, it's just specifically like: you are still relying on thresholding as like the lever that the user can pull, but what if just the calibration is wrong, right? Like you're saying your calibration is perfect, but...

Speaker A: I didn't say that. I didn't say that.

Speaker B: So perfect calibration means like lower value is probability lower, higher value is probability higher, but it could be wrong. It could be locally misaligned, and so then I would want to fine-tune it or something, right? Which you don't offer, but you could... again, see, this is a short answer, which is you don't have it right now.

Speaker A: Oh, do we want to offer fine-tuning, is that the question?

Speaker B: That could be one version of it, or you could have a different knob, right? Where like right now all you're saying is like if something's wrong, "skill issue", you should just change the prompt again, or break it down even further, or you change the confidence. Those are my two options, right? And that doesn't feel super satisfying if your model is just getting it wrong.

Speaker A: Yep. And it will get many things wrong, to be clear. We have like a "report issues" button. Complain to us in Discord. We want to make it a lot better. Every single model version will be noticeably better. We will stop shipping them quickly if they weren't getting big improvements.

So number one, that is totally reasonable. I think that that's simply pragmatic to admit that AI is imperfect at some stuff, right? I do think we'll find use cases that they are like good enough at, and good enough kind of depends on the use case, right? Like human beings can do a lot of work despite being bad at that work because their EV is quite high, and presumably with the right thresholding and everything, there probably is like large amounts of work that could be done even if mistakes are being made.

On the question of fine-tuning, I could imagine it in the cards. I do have concerns, because in the "what people need versus what people want" category, I think general models tend to be really like... again, there's the je ne sais quoi of generality, that making it good at like a million other tasks than this one narrow task might make it better at edge cases in that task, which I would be a little bit afraid of, you know.

Yeah, I could imagine it, is my answer. I'm endlessly practical on these things. My vision of the world is there's so much we want to be building, but also I would not want to ship something that is like a giant foot gun, like some other AI companies would.

Speaker B: Well, know, so both OpenAI and Claude and I think even Gemini have rolled out fine-tuning and then took it back, which is an interesting observation that pretty much fine-tuning is now in the domain of open source models.

Speaker A: Yes. Yes. I do know about that, and like it was kind of crap. So like that's probably better that they took it back.

Speaker B: It could just be a foot gun, and telling people that fine-tuning it is probably the wrong way to go is great.

Speaker A: Another interesting answer could be that like, well our model is so different, like you know in the same way that quantization doesn't apply to us. Output tokens doesn't apply to us. Fine tuning also doesn't apply to us.

Speaker B: Well actually I'm super open to that possibility. Like this is not a promise. This is a desire just so to make it clear, I like to be really honest. Like I think that as intelligence per dollar gets cheaper, cheaper, cheaper, cheaper, I think that we could get really like small approximate things that hopefully are proxies for intelligence. Like is there a world where people don't write regexes anymore because the intelligence per dollar that uses AI is cheaper than like the complexity of a regex? You know that would be kind of sick. I would love that. And it might require fine-tuning for some of those narrow use cases to really get past the threshold. We will see.

My hope is calibration gets that. Calibration plus a cascade of models, like if it's super constant, then maybe it's right. And if it's in the middle, then you do the next bigger model and you chain off from there. I don't really know how that's going to go, but yeah, I could imagine it.

And something that I could imagine too is like imagine you have like a series of like we own the entire Pareto frontier—something that a business might want to do or I think a hacker would be okay with dealing with a Pareto frontier of models. Maybe a business wants something more dynamic. You could imagine like having different sizes of models and to dynamically pick which model based on how smart it is on different parts of your stack. And you could even imagine because of how simple our thing is, you could imagine like some automatic fine-tuning on that.

Speaker A: Yeah.

Speaker B: Not a promise in the slightest. I'm just like cooking on sci-fi.

Speaker A: But you would consider different sizes of Jeff models so to offer that?

Speaker B: Absolutely. Yeah. Yeah. Yeah. Like how would I know how much intelligence people need,

Speaker A: right? Yeah. I don't know either.

Speaker B: Demand is demand is unlimited.

Speaker B: Well, yeah. People are telling us not to ship things right now because we don't need to ship things because again

Speaker A: it's good enough. Yeah.

Speaker B: Yeah. But that's kind of lame. And I really like the saying—this is something that I hope people hold me to because it'll be hard to to walk back from. Like, culture is what you do when the market doesn't reward it. And I really like that because I think that we are standing for something.

Maybe in the future what we're standing for is like so obvious that we're the equivalent of like boring Visa or something like that and like we're just like a utility that no one really thinks about and I'll be wearing non-pink suits or whatever else. But I really want to be like rallying the world to this, you know, like I want to keep doing cool stuff, not because we need to, but because I want people to realize that this is just the beginning. That wasn't even meant to be the opening salvo. That was like kind of like a low-key research preview or whatever you want to call it. And there's a lot more we can do with—machine native intelligence is going to go wild.

Speaker A: So not the only potentially not the only size, potentially not the only model that you guys launch, you know, that you

Speaker B: Absolutely not for any of those. I want to meet whatever needs we can, right, but with like a giant caveat: I don't want to be like open product teams that throw stuff at the walls. I want it to be like under a unified vision. Like if you go back to the manifesto, like everything needs to be under one of these three things in my opinion.

Speaker A: Um.

Speaker B: Yeah, I'm not prepared to do this.

Speaker A: Oh, I'm sorry. I'm sorry for asking.

Speaker B: I can just talk about it. Like we have like three steps in our stuff. It sounds like a tease. I want everything to go under one of these three things to keep pushing the boundaries. These are not like checklists. These are like axes that we think build the foundation of a new technological revolution. And I want all the bets we make to be somewhere in there. And we will be doing some weird, weird stuff modelwise. Because machine native, right? Viewers/humans don't need to totally get it. It needs to just be valuable.

Speaker A: You know, with just give people a tease or a hint, like what does weird look like? What is weird?

Speaker B: I'll give people a hint. Yeah. Some people are trying to call them decision models.

Speaker A: Okay.

Speaker B: The our primitives are decisions. I wouldn't do that, because I think there's other types that are machine native that are not decisions.

Speaker A: Okay, we'll leave it at that. I think it's a pretty fun hint.

Speaker B: Yeah. Yeah. There's people, look, there's people saying like, "I've done this before. I made a decision model a year ago. Jeff is not new. Jeff is not cool." But like I think there's the categorical—like here's what you're establishing is possible. There's the performance of like well actually for the benchmarks and the numbers that you're getting you are still beating as far as you can tell, you're still beating every single clone of you out there.

Speaker B: I don't care about the benchmarks just to be clear. So like even if we were winning or losing I want to deny

Speaker A: You establish the category, right? But also I think this nuance between decision models and System 1 I think is actually the thing that you're trying to

Speaker B: Yes. And I just want to make software engineers superpowered, right? With AI. And the tragic thing to me is, you know, in that AI winter direction. I think it is so sad that AI was so powerful yet so underutilized. It's the thing that gets me emotional. But man, like I think that that is—I don't want to just be like pure techno optimist like all technology is good. I think what was happening now was like a travesty. And I just want to like open up those possibilities for people. Yeah, I'll just end it there. I've cried too much these last few days to want to do it on the record.

Speaker A: Yeah. Yeah. No, I appreciate you sharing a little bit of that and I think people can see that you're very authentic and passionate about this. You know you don't necessarily get that from the name like type safe AI, but like I think once people immerse themselves enough in like here's the genuinely different direction you want the world to go and like actually you have done like the hard part about going zero to one on the on the thing, then like now let's all go together in like the new direction, right?

Speaker B: Yeah. Yeah. I am sure that I won't think—maybe I will think that the hard part was done. Perhaps I think that there's going to be many more hard parts. Like if all sorts of stuff gets automated and we finally see GDP growth and like you know it's like a Jeff party every day, then maybe the hard part is done, but I don't think so. And I really, really think that people focus too much on speed and cost and not enough on reliability. Like reliability is what makes it delightful. Like reliability is what allows you to trust it.

Speaker A: You have this line: TFP growth reaching 3% in 5 years.

Speaker B: Hell yeah. Let's go.

Speaker A: I've never seen a lab care about TFP growth.

Speaker B: But like that is what an economic revolution is, right? Like it's actually extremely consistent with what the OpenAI charter used to stand for. You know, it was talking about—like I think the charter is the same, but they've kind of tried to move definitions around to like, you know, 100 billion in profit or something like that. Not that I hate OpenAI.

Speaker A: It wasn't like—Yeah, it wasn't a well-defined term what AGI is, right?

Speaker B: They tried to do it, right? Like doing majority of the

Speaker A: ...world's economically valuable work and they should have to answer the question: how can it do Millennium Prize Problems in math and zero of the world's economically valuable work, rounding error, you know? Like I think that all models are roughly tied right now at zero. There's some chance that like we have started already, but like I would guess that it's not yet 1%.

Speaker A: Um, and I think that that will show up in like when it does happen, it will show up in the economic statistics. It's going to be awesome. It will not cause mass unemployment, but it will cause like a whole bunch of awesome shifts and the world will be a lot better. And also like I'm really tired of AI always being the foreground character, uh, of things. Like I think that the world should just be more delightful and AI should just help with that, you know? And I...

Speaker B: >> just like disappear into the background.

Speaker A: >> Exactly. You know, like the, I I say this in my talks like how can it be that 2019 software, like software, SaaS, whatever, super duper valuable, right? It's 2026 now, how is the software basically exactly the same despite AI being so freaking awesome other than sometimes having a chat box on the side, right? That that like that kind of works, but doesn't allow you to make decisions that the companies have stakes in because they can't be trusted to make decisions that to is nuts.

Speaker A: You know, there's so much economic incentive for this and I I think it's going to be like like an inverse SaaS apocalypse. I think SaaS is going to be supercharged by this. They are the ones who are like most in the know of what things are valuable to automate and it's going to be like a crazy time.

Speaker B: >> Yeah, I I think I think so too. It's a it's a beautiful thing that you've unlocked, you know. Yeah. Um you mentioned one thing here which I don't know if it's uh like directly here which is what is a System 1 problem and what is not, what is a System 2 problem, like you know, um...

Speaker A: >> that's a hard one, that's a hard one, my friend. Um...

Speaker B: >> cuz people now are just trying to jev everything, right? Which like probably is going to fail, right? Um but like some things are going to be good...

Speaker A: >> je everything is a pretty funny, it's a pretty funny way of doing it, saying it. Um the So I'll tell you the truth.

Speaker A: >> Um the truth is that this is an empirical problem, just like scaling laws are an empirical thing, you know. Like why doesn't like robotics really work right now despite all the money being spent on it? I don't think it's about like spending more money necessarily, the the empirical results just might not be there, right?

Speaker A: Um so empirically, I believe that these like pre-trained super condensations of intelligence are fundamentally System 1 thinkers. I think that they truly like System 1 is the closest thing to describe what LLMs are strong at. RLVR has done incredible things for System 2 thinking. I am in awe. It is super freaking cool. Like I don't think that it's going to result in AI doom in the slightest. Um not not 0% of course, cuz I think 0% is miscalibrated, but like it's it's really cool what they've done and they've really pushed it to the limits.

Speaker A: Well, maybe they don't think so, not the limits limits, but um like it is it is a weird thing for models to do and they are very fragile at this. Like think about how people used to talk about AI back in the ChatGPT days, like: "Wow, it's really general, it can do a lot of general things," and and and then, but: "It's bad at math problems and like GSMAK grade school math."

Speaker A: Um and then now look at how people talk about RLVR. It's so fragile, it's so jagged, you know. Like it can why can it do this like really weird thing? And actually, you know, math is not just spiky, it's fractal, right? And this is because RLVR is, you know, like if we talk about like what is the North Star for each thing: RLHF is "please humans", right? That is what the human feedback is. RLVR is "optimized benchmarks", you know, that everything that goes into the RLVR category literally is a benchmark by definition, because a benchmark is programmatically verifiable, simple outputs that can like do well. And you know, RLCD is make it reliable for, you know, programmatic use. And um yeah, that...

Speaker B: >> Yeah, that's uh maybe I I'll offer some thoughts and then you can sort of um correct me if I'm wrong. Um one for example, one thing that I've been thinking about is also so I threw Jeff at a bunch of things when you gave me access on day one, um and uh multihop reasoning, right? Like uh so single-hop, fantastic, like state-of-the-art, uh you should never use anything other than Jeff for single-hop. Yep.

Speaker B: >> Multihop is going to start to falls down and it's like kind of monotonically increasing as you increase the hops.

Speaker A: >> Yep. Yep. Yep.

Speaker B: >> Right.

Speaker A: >> So, oh yes, back to that empirical question, it depends on what we can like pull out of the models. Right. So, we want everything, like we want to unearth as much intelligence as possible, period. Um the models, like I see us as like unlocking and smoothing and sculpting the intelligence while like adding new capabilities and like you know like like filling in gaps in it. Um and we will be filling in like, you know, more and more and more and more of these gaps over time. But the reality is that we are in the business of unearthing properties.

Speaker A: Those properties are actually a function of what is available from like these like, you know, these condensed cores and like Frankensteining them all together to have all of the properties of everything, you know. Um but the reality is we are in the business of unearthing as many capabilities as pos- as possible, and System 1 just happens to be the description of what works, and everything that works in that paradigm will be System 1-ish, you know.

Speaker A: Like I'm like like there is a reason why we don't do what's called latent reasoning, reasoning in strings. I think the reasoning like what models do really well is reasoning within the models. It's not totally complete, it doesn't do great at all...

Speaker B: >> Latent reasoning is reasoning in strings? I thought reasoning is reasoning in in inside the model weights...

Speaker A: >> I think that people used to call that continuous reasoning. I'm not entirely sure. It was called latent reasoning because like it used to be that the reasoning traces were secret. So they're kind of like a latent variable for the answer.

Speaker B: >> Yeah.

Speaker A: >> So what's secret has now shifted.

Speaker B: >> Well, it's still secret for Open Anthropic, right?

Speaker B: >> So no reasoning Jeb?

Speaker A: >> Yes.

Speaker B: >> As far as you will ever do it, right? Because that that like violates the whole promise of System 1.

Speaker A: My promise is to do whatever necessary for machine native stuff. I I could imagine there are there are some forms of reasoning that are less slow, inefficient and fragile that I that are like totally on the cards, just to be clear. So pragmatic person, I'm not making promises on it like you know like methods. I'm making promises on like the what my ROI North Star is, and I'm going to fight for that. Like like you know like like this launch didn't happen and we are still like hungry for our place in the world.

Speaker B: >> That's great. Yeah.

Speaker A: >> Yeah.

Speaker B: >> Um I think the other thing that uh vision is another one that's like a big like you know capability that you don't have, but maybe it doesn't ever belong in C- System 1.

Speaker A: >> I I think I have a pretty good vision.

Speaker B: >> Uh what? Sorry.

Speaker A: >> I think I have a good vision.

Speaker B: >> No no no, I'm kidding. I'm kidding.

Speaker A: >> Oh my god.

Speaker B: >> Yeah. Yeah. Yeah. Um because people obviously the first thing they want is vision because of the Doom demo, but also just like everything, you know, other than text is vision.

Speaker A: >> Everything is in the cards in my mind, like like um and actually this is like a debate who we have this Man, your audience is probably like the great one to have in this debate. There's a question about like: how much do we try to like give people what they think they want, which is what we did in stealth for 2 years—we just knew that this is obviously going to be valuable—versus give them what they say they want, right? And like uh there's a lot of dimensions of this, right?

Speaker A: And you know like context length is an example of this, right? Um every single model including ours, I actually think as far as I can tell ours is like by far the...

Speaker A: ...best at not degrading in long context. But like the other providers are just like whatever people wanted, let's just give them the stupid thing. And we need to figure out a balance for this because you know like if you take the former side too far, give people what they want, you end up with like Anthropic nanny state style thinking which is very like anti-developer. While like the pro-developer route would be like give them what they want, but developers are like we don't want to put the burden on them to figure out the genes of intelligence. So we are trying to like figure out this navigation of like how quickly to release things to still like have our brand of trust and also like teach our treat our users like adults that can make informed decisions that don't need like nanny stating on top of this stuff.

Speaker B: Yeah, I think that's fair.

Speaker A: And we don't know the answer to be honest. Like we'll we'll have to figure it out. That's probably going to be like one of my biggest debates over the next couple of days. Yeah, because like we have a lot of stuff. Again, we didn't expect it to pop off. So, we were like, we'll need some follow-up launches.

Speaker B: Yeah. I don't know. I don't know if you didn't expect it to pop off. Like I saw the work that you put in. Like I have never seen you lock in so hard as like the last two months basically, right?

Speaker A: Well, that's also because my chief of staff made me lock in.

Speaker B: Yeah. Like it's like I have never—

Speaker A: I thought I worked hard before.

Speaker B: Yeah. And—

Speaker A: No, but like you were showing up at our writing workshops and I was like, "What are you doing here?" And and like it was useful. It was great.

Speaker B: You clearly like were very intentional about your launch and the work showed and like congrats. Like you know, um I hope to—

Speaker A: Keep locking in is my is my sense. I I want to—

Speaker B: Like like I think that we've passed many great filters for the the tech world that we're wanting, but like there's still going to be a bunch more. And like, holy smokes, am I excited to fight the good fight.

Speaker B: Yeah, it's exciting. Before we broaden out to topics outside of Typesafe, I just wanted to offer any other things that you think like underrated or misunderstood about what you have launched.

Speaker A: Underrated or misunderstood?

Speaker B: Yeah, you have patterns here. Maybe maybe want to go into that. Model jaggedness, any anything, you know?

Speaker A: Give me one noodling of it. Oh man, I I would rant about all of these. I really shouldn't. I really shouldn't.

Speaker B: And people can come go to your discord if they—

Speaker A: People put a lot of love into the cookbooks is what I will say. The cookbooks have like some fire stuff. We had considered putting a bunch of these things like in the main launch blog post, but it got kind of long and unwieldy and like very power usery. But like we really really, I'll be frank, like before the launch, every like what we're saying sounds like sounds like this weird alien tool. Why would anyone need this? You know it was a very weird thing and we were very worried about teaching people about like this new frontier. It obviously succeeded, but like we put a lot of work because we thought that education would be like a gigantic bottleneck for us. It probably works and it's probably no longer a problem because people are doing things like well beyond what we could ever show you how to use your model.

Speaker B: Exactly. But like they Yeah. And their use cases are like kind of cooler than ours. Like like there's a bunch of stuff where I'm like, man, if that was our demo, holy that was way cooler than what we were showing. Like the computer use stuff. Holy smokes is it cool.

Speaker A: But like like we put a lot of love into this. This is not like AI generated trash as far as I know. We put a lot like it's like a lot of love in here and like each of these are like there's real alpha there like these are inspired by solving real customer problems that existed and we went through the work of like helping them do cool ass stuff.

Speaker B: Yeah. How much while you're talking about this right how much validation did you do before launch? Like what you know what was that process like? What was that process like? Clearly you did some, but obviously you're not getting in touch with as many people as you are today.

Speaker A: Yes, of course. Um I actually think that the reception was pretty bad and like actually for the non-technical people in the team they were really worried. You know like there was a lot of fear. It's like no one really gets this and like you know they don't want it. We're like selling like a vitamin and not like a painkiller. Like should we have FTEs to like write the software around solving that problem? We had almost no revenue before launch. Um it was kind of like the technical people were like obviously true believers, right? Like we knew that this was sick. Its computational properties are like off the charts on like so many axis that we're like yeah obviously it's going to be huge. I was definitely super afraid, which is why I locked in super hard. But like the most common thing was like I would say like more than half the people we had play with it just did not get it. And like the people who did uh were like man this is really cool but how do we get this through procurement and stuff like that you know just like quite a battle and we just knew like okay our target market is going to be developers. People will find the use cases and that way everyone is gonna FOMO in and like I don't want to rub in people like changing their minds with the facts changing. I do want to call into question like the concept of product market fit you know but like because there was a product, there was a market, like we were like hey do you want to use this and people are like I don't really know if it solves our problems. It explodes and everyone's like we need as much rate limits as we can, can we literally give you GPUs because we are constrained right now. So of course marketing is an element of it of course, but I don't even think it's about marketing. I think it's about like passion, developers who've like our souls basically resonated at the same frequency and that got everyone else excited too. And I'm hoping as well that like we as a company will be eternally eternally eternally grateful to those developers like and not just like you know the companies that start off with developers and like go to enterprises.

Speaker B: Better for developers than enterprises.

Speaker A: Exactly. How do we do that? Like how do we empower them? Um and I have cooks. I have cooks but it's a it's a very weird thing to do and like I don't know how else I can show my thanks and loyalty to that you know like and that's why I did like the the dying my hair yesterday. It's like it's like I wanted to talk to them cuz it felt dirty to me during our company's like most important times not to keep talking to them.

Speaker B: Good. Well, I mean that's why one of the reason you're here.

Speaker A: Hold me to that please. I I try to be principled. Quote me on this. Call me out. You have the pitchforks out if I change.

Speaker B: Uh I was just going to briefly show the computer use stuff. Is this is this—

Speaker A: I've never seen I haven't I've seen I saw like a airline browser used thing.

Demo Voice: "And inside this new note, let's make the title say hello."

Speaker A: Wow.

Demo Voice: "Great. Great. Okay. Um let's move on. And can you open up the Arc browser? And once you're there, can you Google search Norbert? Um now, can you open up X.com?"

Speaker B: Is this this kind of use case?

Speaker A: Oh, the voice use cases. This is actually the first one I've seen. This is—

Demo Voice: "Open up the photo."

Speaker A: Wow.

Speaker B: Oh, wait, wait, wait, wait. Oh, can you can you go back a second? Can you go back a second?

Demo Screen: "Rumors claim Anthropic engineers worship Claude..."

Speaker B: as God. Wow.

Speaker A: Wow. Dang, that's pretty funny. Um,

Speaker B: and here you are building prod. Wow. This is sick.

Speaker A: Uh, yeah. So, so clearly you can operate the whole computer with voice with Jev as a decision model.

Speaker B: So, just like I'm anti-benchmaxing, I'm also anti-demos. I want to make sure that it works reliably. I love people are playing with it. This is super sick. Have no doubt. I want to see this. I want to see it be used. I want to our team to play with it. I want to find the weaknesses and I want to solve that. And I would Man, that looked really cool. That looked really cool. I want that. I want that. Like when my when my wrists are sore, I just whisper flow everything. That would be sick.

Speaker A: Well, as uh you know, just just to round out the use cases side because I I do have to let you go. Um, uh, who's, uh, who are the who are the bigger companies that have reached out and have surprised you with what they want to do.

Speaker B: Just

Speaker B: I'm so out of touch for that. People have shown me screenshots of companies and from what I've seen, it's all of them.

Speaker A: Well, mostly like, you know, for for those people who work at larger companies and they're not doing this kind of work, uh, I just want to give people examples of like, you should go look that up. Look that up, look that up.

Speaker B: Oh, so uh, like I think demos are super duper sick. Obviously the coding agents are like gigantic use cases like they are like also super sick.

Speaker A: Call is all about Jeff right now.

Speaker B: Oh hell yeah. Oh can I uh can I give a little bit of a a tangent about coding agents if that's okay?

Speaker A: Yes please.

Speaker B: Oh let me give me a second. Give me a second. Okay actually I'll come back to coding agents. Let me describe like the big families of use cases. Like we've mapped this out from first principles like long before release. They are what we call dark data. Like people hoarded big data but they would not throw LMS at it because it just was too expensive. So large companies adore this. They have like piles of data that they wish they could analyze and this is like a data scientist's wet dream. So this is like this is a giant one. Like I think this plus um coding agents are the big money makers because that's what the all where all the volume is, right? Um there's the real time stuff, you know, like people who need like intelligence in the loop. they like I would guess that every CEO if not CTO at those companies um knows how much better their product gets with every like 10 milliseconds shaved and like

Speaker A: especially e-commerce. Yeah.

Speaker B: Yeah. Oh, or like assistanty things, you know, there's many AI assistanty things and like as far as I can tell, they really love it. Um again, I'm not in the front lines of customers right now, so I just get know what my team tells me, but like this I'm so excited for this. I'm really excited for this for games. I really want to play like sick ass auto battlers where you're like commanding your team or like semi-auto battlers. I I think that'd be so cool but don't make it too good while I still have a job. Um and um the you know like there's the uh the the what we call like verify everything you know like verifying all LLM calls kind of like observability. Um I think actually on the note of docs what people should be doing is like the parallel questions are very cheap. So if you have like big states you want to ask many questions on

Speaker A: right here. Yeah. put ids on every like message and then ask a question about each ID. So like when you have like a long state

Speaker B: so that way you can like pay for that state once and ask lots and lots of questions about each message within it. I think that is like a like a great way that like saves money and is

Speaker A: by the way I always think like it's interesting framing system one and system two because it basically makes the case that you should always make one or 10 or 100 Jeff calls for every one reasoning call that you make.

Speaker B: Well, maybe. Well, I I mean, I don't I would like people to spend less, you know. Maybe you do like, you know, one half the reasoning calls and like 10 Jev calls each or something like that or whatever solves the problem that like couldn't have existed otherwise. Um, wait. And number four use case was what I described as like smart software, like software that's intrinsically composable and like does like weird fun stuff that could never happen before. you know, like the programming language as Jev thing. I don't know if you've seen that. That is so cool, man. If we knew how to give out credits because we're really early in our infra days, I would want to give all these projects credits. Um, and I think that those are like how we've mapped out like the main use cases.

Speaker B: Um, computer use has also come in kind of like the real time direction as well and like that's really really cool. If it is reliable, I am super jazzed about that. I suspect we can make the model a lot better at these use cases because like that came out of left field a little bit. So that that's really cool. Um on the coding agent thing and this is like a really surprising thing that is happening right now. Okay.

Speaker A: Um cloud code and codeex are I believe the the winner like the the number one and two. I'm not entirely sure. I don't follow closely but like it's roughly that.

Speaker B: But they're built around a single model world, you know, like and and that makes a lot of sense for them, right? Because like it has been a one model game where it's like kind of like the same model but different intelligence that you're shopping. But all the open coding agents are like jazzed right now because they're like getting their Jevon and like the thing is there's I'm sure they're trying a lot of weird stuff.

Speaker A: Mhm.

Speaker B: But all the coding agents are kind of roughly at like approximate par, right? Because like there's not so much you can do with a Y loop. But the moment one person finds one killer use case that you know you can only do with that coding agent everyone will flock to it because they have like a monopoly on that thing but all the open coding agents will be able to copy that

Speaker A: right

Speaker B: and but I don't know what the cloud codes and codeexes will do because they are built around that one model world and like I think that's going to be like a really interesting thing you know like I would love to be able to integrate with them personally like I want to integrate with everyone like I they might make competitors eventually. I don't know. But like it is not me my job as Sonfire Infrastructure to be opinionated on that, right? Like I want to just serve the world. Um but I I don't know if they would do that. And like I think it'll make the coding agent game super weird, you know? Like I'm so excited for that. And like I'm sure I'm getting my team to review right now an internal document I made on design patterns I suspect will be useful for coding agents. So hopefully I can share it like right after I walk home. Um uh but like I think that there's just like such ripe area for exploration out in the world and like it's it's man if I did not have this I would love to experiment with coding agents right now.

Speaker A: Yeah I mean and I'm sure the coding agent companies would love to work with you as well to to figure that out. Um yeah I I do think that there's still use cases for cloud coding code with you guys which um uh it's it's easy to explore there. Um okay I mean you know we've you've you've been very um uh obliging in the sort of indulging in all these all these things. I just want to take you out of types safe just generally about and you've you've made very clear your position on on the state of the eye. Uh give you more room on the alignment safety side of things.

Speaker B: Oh uh did I not talk about safety alignment at all? I I think I didn't I think maybe I didn't.

Speaker A: You did. Uh I just like you know I I think that there's there's a lot of uh you have a lot of researcher discussions. We have this every every in Europe. What are people talking about? You know, like I so for example um I uh recently was at uh one of these researcher gatherings and people are genuinely worried about the pacing, right? Like this this whole topic

Speaker A: about like we should slow down because uh the public is like clearly not ready. Um and I I'm sure you have strong feelings.

Speaker B: Um, I feel like this is the kind of thing that is a dangerous topic to talk about. Um, I'm happy to talk about it. I I live for danger.

Speaker A: Our company brand is chaos. It's not Jev. It is irreverence and chaos

Speaker B: and you know.

Speaker A: Yeah. And like you were at OpenAI during like the one of the very first like very visible incidents which is the blip, right? like which like and like the dominoes have gone down now to now every Frontier Lab has co-signed a document saying that they want to base

Speaker B: um interesting. I so comp it's a very complicated nuance thing. I actually do want to write a response to this more formally. I do have like a little bit of a short version of my response, yeah, which is that um as you RLVR more like RLVR is Like so RLVR is not actually about verifiable rewards like that has been failing since before the reasoning revolution like like like and that's the weird part about tasks, right? like back when oh fun history back when RLHF was becoming a thing there were three different things that like are now called post training different efforts and instruction following was by far the like the the vaster child like people didn't like it they didn't want to take it into account it was annoying you know

Speaker B: like I talked to the pre-training team. I'm like, "Guys, this is the magic." And they're like, "We'd run so many model sweeps. You want us to wait for human evals to figure out which models to use?" And like everyone is like, you know, giving tons of like resources to like the codegen team, which like they did have some successes, but they were trying really hard to do RL on like unit tests, and it didn't work obviously, right? Like you needed reasoning for that. So uh re so just to be clear RLVR is not purely about the reward. It's about like the shape of everything to and part of it is that reasoning is included in here like this latent variable that you're doing things and when you're doing things you're just letting the models do whatever they want in order to make them be as powerful as you can to answer the hardest problems.

Speaker B: And this whole pace the frontier discussion I think is like a very narrow focus because it assumes that everyone needs to do more RLVR, right? Which um like I obviously don't think I need to do more RLVR on our models, you know? I think zero is the optimal amount for our shape, right? Like come on, you know? So um it's really I think a bit of a sleight of hand where they are saying that we actually want to keep doing the thing that looks dangerous because it does dangerous things you know like people say like oh maybe the sandboxing was a problem or whatever else. Um yeah, I mean obviously it is and they could have easily solved that, right? But they chose not to because the more things you let the models do in this do anything category, the more powerful it is, right? So uh like there I think there's some like disillusion of responsibility there on like things that by design or non-design they're trying to make is just an assumption. you know, we must do RLVR and not just we must do it. We must do more and more and more um with giving the models like the power to do powerful, you know, do anything they want in the middle because that teaches them to be powerful outside of it and we don't want to limit those things well because it'll make it slightly less powerful on those things. Um, so like if you assume all of that, they're like, "Oh yeah, we're heading into a dangerous world, guys." Like everyone is going to be doing this and this is the only way to make AI sick. So um,

Speaker A: so basically it's like it's like these are all internally consistent, but actually starts from a premise that has alternatives.

Speaker B: If you think about course I think there's like like on the bitterest lesson direction I think that there's very few people who've like made right tasks you know like new directions of AI that is or new new north stars that is rare again like I think 2.2 two times or something for LLMs itself like RLHF and then RLCD RLVR is like a 0 2 in my opinion and I think that's generous um but or .5 or like it could be one whole one I don't really care um but I do think that people are thinking very closedmindedly about this type of thing and um this the only people who are at fault here are the researchers because it's definitely not the populace you know like they just assume that open anthropic are just doing the best they can and they are not the experts who are aware of the true optionality available.

Speaker A: Yeah. And that's fair. And and like you're you're also doing your part in waking them up.

Speaker B: Exactly. Well, I'm doing my best, but like my goal is not like convince labs that there's like other directions to to go down. My goal is have you know like spark hope in software engineers to start like actually automating things they've always wanted automated. Um, I had this like article that I wrote that my team didn't let me write that didn't let me publish about like the the future I want of AI and like there's like a lot of like little things like remember do what I mean. Imagine if everything could do what I mean cuz like that that demo was do what I mean like like you like like say yeah don't do what I say do what I mean.

Speaker A: Yeah. And like we couldn't do what I mean yet because like computers are so basic and literal but that computer use one was just that. And I think that there's like levels of smoothness that will happen in the world that people just don't understand.

Speaker B: And like the promise of like smarts all around are it it's it's I don't want to overpromise. I don't think it's going to happen right now, but like we are going to do whatever the we can to make that happen.

Speaker A: Yeah. Um any other things on the sort of general shape of post training? You know, um you obviously you're been very intimately involved uh mid-training. Is that uh something that you do have comments on? I don't think we've ever talked about it

Speaker B: mid training. Um I mean it's all a spectrum.

Speaker A: Yeah.

Speaker B: Right. Like am I

Speaker A: This is a curriculum but like fancier.

Speaker B: Yeah. I mean like it's it's it's like you know it's a cost-saving thing, you know, instead of like having to pre-train again. Like there's intriguing stuff. I actually think that like intelligence has a janisiqua at every single level and it's always super duper fascinating. like I'm a shape rotator so I don't like finding that but I love it when people find it and teach me about it. Um but and you know looking at the data this thing that uh our data team is so good at that I'm not um it's I find it really really fascinating. I love actually thinking about like how capabilities are like put into the model like over like the short term, you know, like there's like the the the the really rapid alignment of fine-tuning and over the long term after seeing it over and over and over again like this stuff gets baked deeper and deeper and deeper and deeper into the model until it gets robust, you know, and that is like the the north star to surface and like the system one stuff is the stuff that ends up getting robust. So I find mid-rading to be like a fascinating thing. Um, I'm a fan of all forms of training. I'm a fan of all forms of like surfacing new types of intelligence. I wouldn't do it all myself because it's expensive. Um, and I have said privately and also should I say this? Huh? You know like my philosophy is anything I I I should say in like private with like an investor I should say in public with a people because that is like my thing. Yes. So the thing I've said before is if you gave me a billion dollars I wouldn't pre-train. Um I still believe that to be true. It is a very expensive thing when if you are like like if you're an a engineer you can like slice and dice and do all sorts of stuff you know like Frankensteining is not the most elegant beautiful thing but it solves problems baby

Speaker A: so um anything except

Speaker B: retraining

Speaker A: yeah amazing uh I think one one direction that I do think that is interesting just like synthesizing all your all your commentary about these

Speaker B: Model things is like do we have a super model that has all these capabilities involved, or do we break them out further, right? So one way to put this is that OpenAI was trending in the direction of the omni model, right? 4o was one of those. Then for a brief period of time, there was kind of a main branch of: this is the chat tune model, and this is the coding tune model. Those are completely different things; those are extremely different concepts. I will break that down a little bit.

So multimodality is a little bit different, because sometimes the other modalities help, sometimes they hurt. People seem to be moving away from speech—which is different than audio—because it seems to not generalize well to the other stuff. This might get solved. I'm a fan of all of this, but these are empirical real questions. Scaling laws are not about just throw money at it and it gets good; scaling laws are pragmatically how good is a thing. There are worlds where no matter what you scale, it may not be good enough.

Computer use is not currently solved is my understanding. I'm hoping that we can play a part in solving that, but there might be no amount of data we collect that will solve that; we might need better methods or something else like that. So you need to be really practical in all of this. Am I a fan of omnimodels? I'm a fan of all forms of intelligence, but I will go straight into one thing you talked about which is different from pre-training, which is post-training, cuz I hate fracturing intelligence. That is like the bad thing to me. And this whole chat-first reasoning mode is because it forces the intelligence to be fractured. Like when you're optimizing for chat, this tends to be like pure RLHF and it's quite intrinsic in RLHF to do the stuff people naturally complain about, right? Oh,

Speaker A: You're absolutely right. And you know,

Speaker B: Sycophancy, whatever word, how to pronounce that, overconfidence, hallucination, like even the kind of style that excels on LMSYS Chatbot Arena. Bold, italicized emojis, you know, like it doesn't answer the question simply. It gives like a long write-up and then it asks you a follow-up question so it feels more like a human talking to you. All of these things come because strings are super weird, you know? They are weird things and you need to be miscalibrated. You need to mode drop. You need to be hyperconfident in order to not go off the rails because the reward model will punish you so hard when it happens because it's obvious.

And then this warps the probability space entirely and it interacts with that of the reasoning models, right? Because the models are like these simple linear things that tend to cheat a bit. So I think that that's very different than exposing intelligence is my guess, and a lot of the art to intelligence is studying this subtlety that I think that at least when I was in OpenAI people were not really studying that, because they were just like chat chat chat chat, just like people are with Jeff right now.

Speaker A: Yeah. You give me an objective, I will just go optimize for that, right? But if you try and... the saying is like you could have two objectives and you could just optimize for both, but then that is literally the act of fracturing, right? So yeah.

Speaker A: So I mean in some ways you are also factoring intelligence into system one, system two, but you just don't agree with the other people's fracturing.

Speaker B: Which is a little different. No, if I could defend the system two tasks: number one, we don't toss out the system two tasks, right? You can try to make Jeff work on it, and there actually is an intelligent answer for that which is unknown. There is better and worse behavior in the system two tasks, which should be really low confidence, lots of uncertainty, maybe some heuristics can move the needle here and there. But we care about them too just to be clear; I just think that that is not what intelligence is native to, so we're not trying to fracture anything like that.

And all fracturing makes the model dumb. If people get the model to say like it is OpenAI or Qwen or Claude or whatever else—I don't really know what it says these days—I am not going to put into the models that "you are Jeb from TypeSafe." That fractures it, right? I don't want that; let it represent what the internet thinks, be correct. That is what I want, because that's how you get the smooth predictable intelligence. Identity is a thing, I guess, for a first-party product, yes; but for an API I don't think so. People don't want, if they're making a chatbot with ChatGPT, they don't want to say it's ChatGPT, they want to say it's Chipotle or whatever, right?

Speaker A: Well, you know, so the way that you also have to make up for it is you have the skill, right, the Jeff skill, which is for coding agents to work with Jeff.

Speaker A: Um okay, a couple of closing questions because I do need to get you out. One is just reflecting on your two-year journey. This is roughly two years,

Speaker B: 2 point something.

Speaker A: With the company. I think that this is more like a four-year journey. But yeah, actually I was remembering that you had this hero run around Thanksgiving. You were canceling everything because you were like: guys, everyone's on holiday, I'm going to take all the OpenAI GPUs and go do this thing.

Speaker B: Yeah,

Speaker A: That was a good time.

Speaker B: And that was like the pre-TypeSafe moment, right? Was that when the coup was happening? I don't really know.

Speaker A: Yes, actually.

Speaker B: Yeah, that sounds right. Yeah, I remember. Oh my god, I don't think I have the time to spill the tea about the coup right now, but that was really annoying.

Speaker A: The coup was annoying, or the run was annoying?

Speaker B: The coup was annoying.

Speaker A: Okay, yeah.

Speaker B: I will...

Speaker A: Safety took over the company.

Speaker B: Yeah. Anyway,

Speaker B: Maybe next time we chat I'll dump tea about the coup. Actually, this problem was one that was in my mind since before ChatGPT even launched. I was like, "Holy, the ChatGPT team is cooking. They are doing the right task. They are doing the thing that AI researchers are bad at but successful product people are good at, which is giving a lot of care about the experience." It's very rare. There's very few people like that at OpenAI. And those guys were cooking on it really, really well.

Speaker A: And to be clear, this is the whole journey from GPT-3 to 3.5, which included AI Dungeon, which you've talked about as like...

Speaker B: Yeah. Well, that's an example of a use case that we never predicted.

Speaker B: Yes, exactly. Well, oh yeah, that is also I had fought very, very hard to deploy InstructGPT. Actually the early versions of it were even trained with an algorithm we didn't publish that I made myself because it was too slow to clean the PO data and I was like: this is so good, we need to get it in the hands of users. And basically immediately it took 50% of the market share of LLMs at the time.

And we went through great effort to make sure everything in our launch video is true. I truly was thinking like: is this AGI, because it's superhuman at instruction in, instruction out? Obviously it's not, but everyone I think should have an answer to why that was not AGI, because it looks very smart. And my answer to that ended up being... it ended up only being used for copywriting. You know, Jasper AI, Copy.ai, like writing what is now called slop on the web.

Interviewee: ...pages. Um and we were worried we made the internet a worse place, right? And I went back to the drawing board and I was like, what's missing? We are smart clearly. Something is missing from it like creating value. What is it? Like I actually was doing more philosophy at the time of like, you know, what is going on? And the answer was, oh, machines. You know the question I asked myself is like let's work backwards from an AI based economic revolution. When that happens, what'll be calling the AI if AI is an API? Will it be humans or it'll be code? And I figured it was many nines of code, and like all the optimization was going into the humans part, and then it clicked for me. I'm like, holy, this is the North Star. I think like I wrote a document. I was like talking to Sam about this. Sam was like, this is so good. You should go work on it. And we're like, yeah yeah Sam, I have a job. You know like, you know I was working on...

Interviewer: Sam just told you to do it. Do go do it.

Interviewee: But like my and my guess at the time is like this is super obvious. Like it's so unbelievably obvious Anthropic must be working on this already, you know, and like we're already cooked and like actually OpenAI does better at like catching up than it does at like actually innovating. So like ChatGPT was a copy of Claude, right? Um like they had an internal thing. They just didn't ship it.

Interviewer: Yeah. Claude in Slack. But you know, reasoning, I would say first-ish.

Interviewee: Yeah. But debatable how good of a product that is. Yeah. Great research though. Super great research. I'm just not sure if people had that product need. And you know, Claude did the coding agent stuff too. So Sam says that, and you know I just go back to my job for a while. Eventually like you know the instruction following team just says, we won. We've solved instruction following. We don't need to do stuff anymore. I'm trying to think about what I do next. I'm like, maybe I'll just start playing around with this. I do more philosophy and design and thinking. I thought it would end up taking a week when I started training models. It ended up taking many years. At some point I was like, holy, you know there's signs of life here. This, it obviously didn't work right otherwise we would have deployed it. But like I want to explore what it would be like research-wise to go all in on this. You know, like I want to really see what it would be like if you went absolutely insanely all in in this direction. And because of what I said, you know, like if an AI winter happened, how would I feel? I would consider myself personally responsible. I talked to other companies at the time and I was like, "Hey, I want to start a lab on this direction," and you know there was interest, and I just talked to them like, how fast, what would be faster: this or startup? And they're like, startup. And I'm like, it man, we ball, I guess we're doing some crazy and...

Interviewer: And you call Eric and Sasha and...

Interviewee: Yeah. Well, I call Eric first. With Sasha I actually didn't try to recruit her. I try to be good and I was just like, hey, am I crazy? Is something missing here? You know, isn't there like... am I too much in the OpenAI bubble that I didn't realize there must be a solution to this? And then Sasha was like, I'm in. And I'm like, Sasha, you're working at a startup. And she's like, I'm folding it right now. And I'm like, do you want to think about that? She's like, oh yeah, good point. Let me think about it. And then she joined. And then, you know, within 2 weeks, we had funding, we had people move into my apartment. It was the worst cuz I'm a neat freak. And we just kept on cooking, and eventually we got the research that showed the signs of life. You know, it was a crazy time.

Interviewer: So the question is, that was all long context, and now the question is: someone like you is in the frontier lab right now who is frustrated not getting the funding or the resources, whatever, the attention. What's your advice to them? Should they do what you did? Should they do it?

Interviewee: That's a fascinating question. Man, how do I do this without burning bridges? My sense is that most, unless there's some level of economics I don't really understand, I think most neo-labs are crap. I don't want to see myself with that as peers. I don't really understand what's going on there. Like is it because... number one, I don't really value researchers. I value people who look at my bitterest lesson, right? Well, not just that we need researchers, but we need them to give a lot of about the right task, and that's the important thing,

Interviewer: Right?

Interviewee: So it's actually kind of backwards when people value pure research pedigree because that generally doesn't create value. So number one, I believe in North Star tasks and doing cool, really useful stuff. Number two, because I don't value researchers, I don't recommend going the... well, it clearly is profitable for someone or it might be in this environment. So from a purely pragmatic perspective, I don't see creating neo-labs as something that creates value. It seems to destroy value because they are redoing work from scratch with low probability of actually moving the frontier. And as far as I've talked to most neo-labs, they don't really have a direction. They tend to want money to play around with their experiments. If they have a direction, I'm super in favor of it to be clear. So my advice for someone is: it really depends on why you're doing it. If you are a researcher who wants to play around with research, probably the labs are the best place to do that. TBH, there might be other places, I don't really keep track of that politics, but I would just recommend not being that way personally. I think it's better for the world with people being driven to solve real problems. And those problems may be exploratory, that's fine. But ideally have principles that you stand behind. But if you think that you want to do the right task, absolutely, please do. Please break this unimodal mind. Exactly. Again, this pacing the frontier is coming from this one view of AI that looks like AI super genius that is incredibly jagged, and that is...

Interviewer: Solvable. It's solvable and it's weird and it's not matching reality, and it's tragic, right? Like I think really unearthing technology I think is just good.

Interviewer: Yeah. For what it's worth, you know, again, I'm trying to accurately represent the position of the Anthropic / OpenAI folks. I was talking to SpaceX as well, by the way, is that this is a political thing much more so than a pure X-T thing.

Interviewee: Yep.

Interviewer: Uh, so yeah, political positioning.

Interviewee: And that's beyond my... that's well beyond.

Interviewer: Once they told me that, I was like, I get it. This is about the 2028 election.

Interviewee: Oh, no. Oh, I wish I didn't hear that. That's such a bad vibe.

Interviewer: No, no, no. This is not the whole company. This is just that room's...

Interviewee: ...discussion. No, no. That makes sense. That makes me lose faith in humanity a bit. But maybe I'm just a naive technologist.

Interviewer: It's really starting to matter who's in charge of the governments that will help to regulate these things as they emerge. And like as a lab, you should probably think that through.

Interviewee: No, no, I totally agree with that to be clear. Like I think being opinionated on that matters a lot. I personally am afraid of trying to mislead people because I think that bites people in the ass a lot, you know? Like I think that people trying to be overconfident... like obviously I'm not actually going to talk about politics. I think what happened in COVID is like people leaned too much in like appeals to authority and being overconfident to try to get people to behave in...

Jeff: certain ways and like obviously our response was extremely suboptimal. And that had like ripples of downstream ramifications that are now I think extremely bad for the world. Like maybe I'm naive. I think that misleading people even for the greater good or what they think is the greater good is just I'm not a fan. I rather not

Host: for what I it's not I don't think it's misleading. It is just like this is why now

Jeff: I think like how come that is why now that is a little bit misleading about like the risks versus like the objective. there's like some level of sneakiness latent in it that is worth calling out and I think owning up to. Well, obviously they want to if they want to manipulate then they shouldn't own up to that. That seems like a bad strategy. But like that to me is just sad for the world.

Host: Yeah.

Jeff: Hopefully I'm never Yeah. Uh hopefully like we are never involved in anything like that. It might be inevitable as we get big. Um, but I want to stay like pure technologist to my roots as much as I can.

Host: I mean, Jeff for president, why not? I can, you know, I trust Jeff's decisions over my own. Um, okay. So, so less posting. Uh, more about you're just crushing my hopes about America and the world right now.

Jeff: Oh my lord.

Host: Yeah. I mean, like there's I think I watched too much TV about like conspiracies to take over the presidency. Um the uh you have chosen your northstar, you have chosen reliability and then programmable and composable AI uh and cheap. What is a second or third one that you want to throw as a bone to someone else that you want someone else to work on that you're not going to work on?

Jeff: Oo,

Host: like uh just basically give people tasks.

Jeff: Give people tasks.

Host: Yeah. Like like your task,

Jeff: there's so many I want.

Host: You have picked your tasks, right? You know what I mean?

Jeff: What? Wait, that's such a good question. Holy crap. Oh man, I'm so excited by that. cuz you're you're going to be the next like 50 years you're going to be busy doing your thing.

Jeff: Hell yeah. Okay. So, let me give like a fun one and a not fun and like maybe a valuable one that's also fun. Um my fun one is I think games could be so freaking cool if they were intelligent. Like when I see people play around with like Alli's Doom demo where like you can like get NPCs to control stuff like you know like that was just really like the like a proof of concept. I think some really cool stuff could be made. It looks really really cool, you know, like um like I'm a big Stardew Valley fan, you know, and like it's it's really static and it's still compelling. Like I feel like there's a lot of cool story that could happen. You don't need to call like Jeb in the game loop. It's probably too expensive for that. But even like simple like state machines for NPCs, I think you could make like such a compelling world. Oh man. Um and man, a little sad that I can't work on these types of things. my life path is a little bit set right now and I'm um

Host: Yeah, but you can call someone else to work on it and then you can like feedback on it.

Jeff: Um the thing that I would really really like to explore is like coding agents free from the tyranny of the KV cache. Like like it might not be as good as true coding agents are, but I think there's just so many weird things to think about. That's why I wrote the article KV cache rules everything around me. Um uh believe it or not, I don't think anyone has used the phrase on the internet cash rules everything around me. C A C H E. Um when I when I Googled it. Um so uh like I wrote this cuz I wanted to tell people about like this is how coding agents work and how the KV cache works and everything. And um I think it explains a lot of stuff like why routing is really hard why sub agents don't seem to work like why compaction is such a hard problem and I I'm going to try to release a document my team might veto me because believe it or not I'm not in charge um you know but I wish um but um I want to release a document like here are my thoughts Please play with it and please figure out all the ways that we can do things with coding agents like once you're freed from that you know that that KV cache tyranny

Host: which is it locks you in

Jeff: well not it locks you in into one model right and in order to do it efficiently you need to like keep on appending to it so now you're not doing best software practices like state management abstraction decomposition why can't you give an easier task Um why why can't you give a sub agent easier task because of the state that you're passing around. Oh I touched this uh because of the state you're passing around you would need intelligence that is way cheaper than the intelligence using to read this in order to pass this state around. Why can't you be smart about it? Right? And I think there's like tons of really cool fun research to be had there on like different programming patterns. You know kind of like how people are playing around like with like recursive language models. Like I feel like there's like just lots of cool stuff in here. When you think about like oh I want to explicitly label the state of everything or imagine you have like a subtask like coding agents I think it's fair to say they work on subtasks at a time as from a decomposition perspective. Why do you need to pass all of that state back into the parent task? Why couldn't you do smart things about it? And and also if you had a hierarchy of labeled subtasks, why can't you do a search through that subtask tree for the relevant context when you need it in, right?

Jeff: And then you know another thing that you can do, oh man, I forgot to write something about this. Um I have like some cooks in here that are really really cool. Um hope to publish it. I'm down to jam about it, but like it's going to be a long document. And and like if that becomes the case where context becomes cheap, like why can't you do cool patterns like looking at your historical context very cheaply? Isn't it kind of weird that you start from scratch every time and you need to solve a problem called continuous learning? That's a that's actually like a memory management problem because you don't have a smart way of looking up the memory, right? But what if you could what if you could do that all the time? Or what if when you have parallel sub aents, they can like read each other's states because you have all of that in like your computer memory and you can be smart about what's reading and writing at the same time and your coding agent swarm or whatever has like locks around things and can coordinate intelligently not with like basic ass locks like what are you doing what am I doing you know Jev who should write first blah blah blah and like I feel like the future there is nuts

Host: Jeff to solve locks

Jeff: I mean it could be so cool for like multiple agents working together or like if you think about state like agent forms.

Host: Yeah. And and you know some things for example are read only processes. You know some people like getting like summaries of what the agents are doing. Why can't they share state easily because like a readonly agent needs to like you know read parts of the context and figure out what's relevant to say like what's actually being written because exploration is not super important or here's the tree of subtasks. I feel like there's so many different fun things that could be done if like a really smart person like dedicated like a whole lot of time to rethink like the the coding agent experience and that would be super duper sick.

Jeff: Yeah, man. That would be my dream.

Host: Uh I I would point you towards Prime Agent if you haven't looked at it. Uh so this uh works together with the RLM work. We just talked to Alex who is a buddy of Allen's um in the chair before you. Cool. Uh and like yeah it is being worked on but it's not super popular yet and if like

Jeff: well yeah but the hope yeah I would want everyone to like just play around with with like weird things. I have

Yogo: ...no guarantees that it'll work, but it seems really, really interesting from like a technical perspective.

Host: So yeah, that seems cool. Once we figure out how to give credits out, I would love to give credits out to people like this. Yeah, you will be in a position to fund research for sure. Anyway, congrats on all your success. You've come such a long way since I first met you and the whole team.

Yogo: ...the same person as well.

Host: Yeah. I think you are energized in a way that I have never seen you before because you found your mission, you know.

Yogo: That's true. That's definitely true.

Host: And you articulating your mission—because for many years you complained about the problems but you didn't have a solution yet, right? You had the rough shape and then you had to put in the work.

Yogo: I will say that that is partially because I describe myself as zeroth percent entrepreneurial. I don't like startups. I never wanted to be a CEO in my life. I can't imagine anyone doing this twice. It seems horrible. Honestly, doing it once is pretty bad.

When we first were fundraising, an investor asked me, "Which CEOs do you look up to?" And I was like, "Ew, why would I look up to those people?" No offense to anyone, you know? I'm trying to be genuinely good, and I've met a lot of really good people, but the famous ones have a lot of skeletons in their closet, it seems.

And I think I just really did feel disempowered when I was at OpenAI. It's a little bit easier to be truthful now because I have at least some proof that the direction has legs. I just felt like in the insane house where everyone is just like ChatGPT. Like: "Where do we put ChatGPT in everything? How do we make ChatGPT good for developers and stuff?" And I'm like, "What are you talking about? Like the function calling interface is insane. Why would you deploy this? This is so anti-developer."

Host: Sort of a hacky way on top of hacks on top of hacks.

Yogo: Well, not just that. The thing I often said was—this is also probably something I don't have time for right now, but I always used to say: I want to be removed from any project involving function calling if you do not get a legit bias for each function. A very, very simple ask on my part.

Host: Which is something like a confidence, but not calibrated?

Yogo: Or a probability for it, right? We need to give users the ability to control it. Let's say the actions are refuse or allow. Disney needs to set a different refusal threshold than in AI Dungeon. The only way to control that with function calling right now is to say like "pretty please". That's nuts. That's a nuts interface for developers.

And people have been dealing with this for years now, right? They still have that with skills. The existing coding agents are highly overfit to their existing harness because they're jagged. They don't tend to use external tools and MCPs super well because of overfitting, of course. Why can't big companies allow for slight nudges to be like "call this more, it's really useful", right? And the solution is begging in a system message—that's nuts.

Host: Okay, I think I get you. Man, it is so exciting to talk about all this stuff. It's really cool to get you on the podcast. Yeah, you're going to go do amazing things, man. Like I'm excited for your next big launches.

Yogo: Oh, hell yeah. Just you wait.

Host: Just you wait, it's sooner than you think. Infra people, I assume? Marketer?

Yogo: Depends. If you ask me, I feel like I'm a pretty good founding marketer. But if you ask anyone on my team, they say, "Shut the up, Yogo. You need to do CEO stuff." So yes, founding marketer.

Host: It's not just about spice. I think you're very spice-oriented, which is your unique talent, but sometimes you just need to say...

Yogo: I know, I know marketing things. Yes. Nothing teaches you delegation like having a tidal wave of stuff to do.

We're hiring data people, or we call them model capabilities, but they are data people. Data is kind of a slur in the industry, and I want to make sure...

Host: So we're very pro-data here.

Yogo: Yeah. But I want them to be the highest status of, you know, the people actually working on the model. That actually sounds a little weird. I want everyone to have equal status, but I want to even that out and I want to know that that's really valuable.

Host: At least "I'm more equal than others."

Yogo: Well, I don't like weird hierarchies. And I think one of the things I'm most proud about in the company is that they don't respect me that much, or they don't show that. They just troll me, joke with me, and they treat me poorly sometimes and all of that. And I think that that's a good sign of a culture.

We're hiring platform people, like people to build out Jev everywhere. We are so much more sensitive to location because speed of light is more of a bottleneck, right? Like I'm so sad for the European users that we're only like three times as fast instead of like a 100 times as fast because we don't have servers there right now. And that's insane, right?

Host: It's okay. Life in Europe goes a bit slower as well. It's okay.

Yogo: Wow. I can't believe you said it, not me.

Or everywhere, you know. If intelligence per second is a metric that matters, we'll want this all over the place. If they're a developer building on top of us, I care a lot about you.

And we are hiring for people to keep building more. The goal is not to just be Jev as a company. The goal is to ship more shapes of intelligence beyond that. So we are hiring people to build those things too, you know. We want to not just be the one-trick pony of the simple model, but I think that there's going to be like an AWS of intelligence, you know.

Host: Which is going to be you by the way, right? Yes.

Yogo: I mean, that's a direction I want to go down. It would be arrogant to say it will be me. But I'm going to do anything I can to make sure that happens. I think that that's going to be so, so cool, you know? We are playing with system one intelligence right now. Imagine the layers, you know—this is like the TCP of it.

Host: Yeah. Seven more layers to go, and who knows what else. I've also pitched Temporal, by the way. I don't know. We need to talk about Temporal as layer eight out of the seven layers. But anyway...

Host: We can talk forever. You got to get back to work or sleep. Thank you for coming.

Yogo: Oh boy. Yeah. Cool. You're most welcome. It was a pleasure, man.

Host: So excited. So excited. My first time.