Jev：面向生产而非“上帝”的系统一模型——对话 TypeSafe AI CEO Diogo Almeida

Tickets for AIE NYC now open, and apply for the invite-only AIE CODE. Join us!

We have an unusual relationship with today’s guest: for years since coauthoring the InstructGPT paper, Diogo Almeida had been saying that API-available frontier models have been going down the wrong path, everything from the alignment to refusals to reliability perspectives, that we have dropped every mode other than autoregressive chat-tuned LLMs because of the overwhelming success of ChatGPT.

In a launch video now viewed ~40M times (by comparison, GPT4o was 22M, Fable 5 was 15M, Navier Stokes was 74M, and 6 Astra was 137M), Diogo introduced Jev and it immediately took over the AI timeline — we’ll skip full Jev explainers because your favorite AI influencer/educator has probably already done one. We also collected:

- the official patterns and cookbooks you should see first, from Allie

- Jev usecases

- speed based - games and computer use

- the voice + computer use example we discuss at 1h34 mins

- voice + browser control

- The must not miss Doom demo

- Driving cars in games

- Excalidraw

- virtual try-ons

- “Smart Games”/smart NPCs

- guided responses in text messages

- Jev for coding agents has an official guide

- jev for linting

- compacting tool calls

- reasonable pushback from Theo - Diogo has published a note on the Tyranny of the KV Cache that you should read as a followup after the pod for Jev + coding agents, because of his belief that Cache Rules Everything

- Programming Languages built atop Jev (Diogo’s fave)

- Jev for analytics replay and user journey review

- “dark data”

- entity resolution

- natural language search

- “smart software”

- a core goal of Jev is to “disappear into the background” - eg as unremarkable as regex

- Jev as a judge

- Jev memes

- Jev vs LLM capabiltiies

- blending transformers and classifiers

- about the confidence api

- Jev vs GLiNER (note difference/pushback, agreed, agreed, agreed)

- Jev on trolley problem

- Jev Bush

Instead we’ll focus on what we can uniquely offer — a broader philosophical and mission-based understanding of how and why Jev was created, and what you should expect next in terms of future models from TypeSafe (ReasoningJev?) and what usecases and ideas you should work on vs the 55th low effort clone of Jev’s API or doing a generic JevBench benchmark - something Diogo has rejected publicly.

Why RLCD: Three kinds of RLHF, and why they are ALL the wrong north star

Diogo knows a good deal about RLHF, given that he was on the team that pioneered post-training at OpenAI — and traces the three branches to Christiano et al 2017 (the robot backflip demo), Stiennon et al 2020 (learning to summarize) and his baby, Ouyang et al 2022 (InstructGPT). From there on, every innovation from Function Calling to Structured Outputs to Reasoning felt like a hack on top of the string based, sequence to sequence prediction paradigm. As he mentions on the pod, from 2023-2024 he struggled unsuccessfully, due to both personal and organization underestimation, to train a model that accurately addressed what he saw as the core problem with making LLMs the heart of software: reliability.

Jev’s core innovation is "Reinforcement Learning for Calibrated Decisions”, a novel, unpublished technique that optimizes for “answers with epistemically honest probabilities on System One tasks” rather than human rated feedback (RLHF) — which causes hallucinations, sycophancy, and permanent reliance on humans — or programmatically verifiable outputs with rubrics (RLVR) — which solves Navier Stokes but exacerbates jagged intelligence and doesn’t integrate well with other software.

We’ve talked about the calibration problem before on the pod, but probably the single best place to understand why RLCD became necessary is Diogo’s AIE talk, which discusses why a generation of training helpful AI assistants for humans has impaired them for training models for composable, programmable AI for automation.

At the end he also teases his contrarian opinion on scaling laws - which teases how to build a modern neolab without the billions of dollars the major labs have…

The Bitterest Lesson: Tasks and Data beats Compute

We spend a good amount of time discussing Diogo’s essay on the Bitterest Lesson:

His point is that “You get what you optimize for and the bitterest lesson in ML is that the most important part of it isn’t ML at all.” - and picking the right north star, eg upvoting for user preference vs being integrated into tool calls - makes everything else fall in line.

We’re excited to catch up with a freshly dyed Diogo to discuss:

- Why AI can solve extraordinarily hard problems but still fail to automate basic work

- What System One Models are and why Jev is built for software rather than chat

- RLHF, mode collapse, calibration, and the hidden costs of optimizing for human preferences

- Why refusals become a problem when AI is buried inside software dependencies

- Why TypeSafe rejects public benchmarks and optimizes for intelligence per dollar

- The “bitterest lesson”: why the right task and the right data can matter more than compute

- Why TypeSafe thinks of itself as a data lab rather than a model lab

- RLCD vs. RLHF and RLVR as fundamentally different North Stars for AI

- Why reliability and robustness matter more than simple determinism

- Jev’s programming primitives and how intelligence maps into software control flow

- Why developers should decompose AI workflows into small, measurable decisions

- How structured state replaces giant prompts and system messages

- Why Diogo thinks AI should eventually disappear into the background of software

- The “inverse SaaS-pocalypse” and how AI could supercharge existing software

- System One vs. System Two intelligence and the limits of reasoning models

- Dark data, computer use, real-time intelligence, and Jev’s biggest early use cases

- Why Jev could reshape coding agents built around a single-model architecture

- Why Diogo says he wouldn’t pre-train with $1 billion

- The OpenAI journey that led to TypeSafe and why he thinks many neo-labs are approaching AI incorrectly

- Coding agents beyond the KV cache, shared state, sub-agents, and the multi-agent future

Diogo Almeida

- LinkedIn: https://www.linkedin.com/in/diogomda

- X: https://x.com/CompleteSkeptic

- TypeSafe AI: https://typesafe.ai/

Timestamps

00:00:00 Jev Launch Week and the AI Economic Revolution

00:02:50 What Is Jev? System One Models and Programmable AI

00:05:54 RLHF, Mode Collapse, Calibration, and Yann LeCun

00:10:29 Programmatic AI, Refusals, and Safety Alignment

00:17:21 Why TypeSafe Rejects Public Benchmarks

00:20:43 The Bitterest Lesson: Data, Compute, and the Right Task

00:24:59 RLCD vs. RLHF and RLVR

00:28:42 Why Powerful AI Still Hasn’t Automated the Economy

00:39:55 Reliability, Robustness, and Determinism

00:48:11 Model Versioning, LTS, Speed, and Intelligence per Dollar

00:54:04 Inside Jev’s API and Programming Primitives

00:58:28 How to Build with Jev: Structure, Decomposition, and Small Decisions

01:18:28 The Inverse SaaS-pocalypse and AI Disappearing into Software

01:33:21 Computer Use, Dark Data, and Jev’s Biggest Use Cases

01:38:48 How Jev Could Reshape Coding Agents

01:41:00 AI Safety, Frontier Pacing, and the Limits of RLVR

01:48:03 Why Diogo Wouldn’t Pre-Train with $1 Billion

01:55:19 The OpenAI Story Behind TypeSafe

02:01:41 Why Diogo Thinks Most Neo-Labs Are Getting AI Wrong

02:08:00 Coding Agents Beyond the KV Cache and the Multi-Agent Future

Transcript

Introduction: Jev Launch Week and Developer Momentum

Swyx [00:00:00]: Okay, we’re in the studio. A special occasion because this week, Diogo, 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?

Diogo Almeida [00:00:16]: Emotionally?

Swyx [00:00:17]: Yeah.

Diogo Almeida [00:00:17]: Never been worse. Like, I’m a ragged corpse of a person right now because there’s so much going on, and I’m like a technical CEO, so I have, like, a lot of fires to fight.

Swyx [00:00:29]: Yeah.

Diogo Almeida [00:00:29]: 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, like, carnival house of mirrors, and everyone is just insane and saying the weirdest stuff that doesn’t make sense. And it feels like for just this week, like, I’m on a better in sync with reality and like, oh, people see it now. AI can be so much more than what was once thought.

Diogo Almeida [00:01:06]: And like, yes, we are going to make. Like, an AI-based economic revolution is back on the table, and this is fucking awesome.

Diogo Almeida [00:01:17]: I’m so jazzed the developers get it. It’s, it’s, Yeah, and I want to show my eternal gratitude to the developers and

Swyx [00:01:25]: Yeah.

Diogo Almeida [00:01:26]: I’m so jazzed about the community and everything. It’s so great.

Swyx [00:01:28]: Yeah, you were saying yesterday that you decided to prioritize the town hall and not a bunch of, like, VIP, investor-type people because you wanted to make sure that they are the people that you get your most, attention, right? The engineers, the developers.

Diogo Almeida [00:01:43]: Yeah, it felt a little like, oh man, I’m talking to, like, really important people right now.

Swyx [00:01:47]: Yeah.

Diogo Almeida [00:01:47]: I probably shouldn’t reveal who.

Swyx [00:01:48]: Yeah.

Diogo Almeida [00:01:48]: But it feels a little bit dirty for me to, I’m, like, perhaps overly genuine in things. Like, it feels, like, dirty if, like, in my gigantic calendar event of people to talk to, the community isn’t one of those.

Swyx [00:02:04]: Yeah.

Diogo Almeida [00:02:04]: And actually, in my ideal world, it would be, like, community all the time. I was thinking, “Should I host a town hall while walking to your studio?” And I’m like, “No, that’s too crazy.”

Swyx [00:02:12]: Sure. Yeah. Well, you guys have been hosting town halls on Discord. Discord is now 100,000 people. Your Twitter’s

Diogo Almeida [00:02:19]: I don’t follow these stats.

Swyx [00:02:20]: Yeah.

Diogo Almeida [00:02:20]: So holy shit.

Swyx [00:02:21]: Your Twitter’s blown up. It was, it was really funny ‘cause, like, at AIE, you were like, “Yeah, follow me please,” and then you didn’t, like, provide even your handle.

Diogo Almeida [00:02:29]: I’m a noob. I’m a noob.

Swyx [00:02:29]: You’re such a noob.

Diogo Almeida [00:02:30]: I’m a noob.

Swyx [00:02:31]: But no, but that, like, that’s, like, positive aura that, like

Diogo Almeida [00:02:33]: Cool

Swyx [00:02:33]: You don’t know how to promote yourself.

Diogo Almeida [00:02:35]: Yeah. Someone, like, called me out when I posted, like, “Holy shit, we’re all three twending-- trending topics.” And then they’re like, “That’s a personal feed.”

Swyx [00:02:42]: That’s a personal, yeah.

Diogo Almeida [00:02:43]: And I’m like, “Oh, no.”

Swyx [00:02:44]: Of course, of course it’ll trend to you.

Diogo Almeida [00:02:45]: Cringe. Yeah.

Swyx [00:02:45]: Yes, ‘cause it’s what you clicked on.

Diogo Almeida [00:02:47]: Yeah.

Swyx [00:02:47]: So okay. Let’s, Yeah, so congrats on everything.

What Is Jev? System 1 Models and Intelligence per Dollar

Diogo Almeida [00:02:50]: Thank you.

Swyx [00:02:50]: We’ll talk about more, details as you have them. But let’s, for people who are, like, living under a rock or just want, like, the definitive thing, what is Jev?

Diogo Almeida [00:03:02]: Whew. Let me think about. That’s a hard one.

Swyx [00:03:07]: Okay. And I’m happy to, like, re-ask if you wanna kind of

Diogo Almeida [00:03:09]: No. I’m happy to

Swyx [00:03:10]: Okay

Diogo Almeida [00:03:10]: I’m happy to, like, just jam on it.

Swyx [00:03:12]: Yeah.

Diogo Almeida [00:03:13]: I will say, like, 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 ‘cause ChatGPT can just explain it.

Swyx [00:03:20]: Nice.

Diogo Almeida [00:03:21]: So the way I see it is we new-- need a new class of models. We’re not attached to naming that class of models. Our-- the most accurate name we’ve come up with is System 1 models.

Swyx [00:03:33]: Yeah.

Diogo Almeida [00:03:33]: There will be reasons, but it’s-- there’s a reason why we don’t call them decision models, because, like, they will be. Like, 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.

Swyx [00:03:48]: You should have said low-key research preview.

Diogo Almeida [00:03:52]: It kind of was, right? It kind of was. But we. So there’s a class of models that we describe them as, like, machine-native, System 1, large programmable. I think these are-- is the class of models where the goal is for code to be the consumer. So as opposed to, lar-- pre-trained large language models, which are meant for, like, autocomplete of the internet, or RLHF models, like chatbot instruction-following models, which are meant to, like, reply to text, or RLVR. It’s in a weird gray area with RLHF. Like, these are meant to have things that directly are consumed by code, hence the name type safe. So the thing we really want is to have, like, AI, like, 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 a System 1 model, whatever you want to call it. Jev is meant to be optimized for intelligence per dollar, hence the name Jev.

Swyx [00:05:03]: Jevons Paradox.

Diogo Almeida [00:05:03]: Jevons Paradox, 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 is meant to be. 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 trade-offs. And we are just going all out on that.

Calibration, Mode Collapse, and the Limits of RLHF

Swyx [00:05:31]: Yeah. And to me, like, calibration is one of the new things that people weren’t talking about as much. We’ve done an episode In the past, with Clementine Foreia of Hugging Face, where they were like, “Yeah, actually, y- they’re just.” Or, and this is your whole argument about RLHF, is they’re more collapsing towards what you want to hear the most

Diogo Almeida [00:05:50]: Ooh

Swyx [00:05:50]: Or what is most likely, instead of, like, their own internal confidence about a thing.

Diogo Almeida [00:05:54]: Can I soapbox on that for a second?

Swyx [00:05:56]: Go ahead. Yeah.

Diogo Almeida [00:05:57]: Cool. Like, I’ve been heard that your audience is the most technical, so I actually want to get into that.

Swyx [00:06:02]: Yeah.

Diogo Almeida [00:06:03]: 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.

Swyx [00:06:17]: Mode dropping or mode collapse?

Diogo Almeida [00:06:19]: It’s the same thing.

Swyx [00:06:19]: Is that what you call it?

Diogo Almeida [00:06:20]: It’s the same thing.

Swyx [00:06:20]: All right.

Diogo Almeida [00:06:21]: And I wanna have a blog on this eventually, but I, like, want to tell as many people this as possible ‘cause 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

Swyx [00:06:36]: What about this?

Diogo Almeida [00:06:37]: Well, should I address this now or should I wait and go into mode collapse?

Swyx [00:06:40]: No, later. Go mode, go mode collapse. I don’t know.

Diogo Almeida [00:06:42]: So I actually think that among takes, Yann LeCun’s is among the most accurate. But he has this very famous/infamous slide about,

Swyx [00:06:52]: The cake?

Diogo Almeida [00:06:53]: LLMs are doomed.

Swyx [00:06:54]: Okay.

Diogo Almeida [00:06:54]: Like that one where he, like, has, like, a pie chart with, like, a tiny par-- tiny little thing- and says that as you increase sequence length, the probability of it making an error goes in. 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 you

Swyx [00:07:21]: What’s the disconnect, right?

Diogo Almeida [00:07:22]: Exactly. And may I or you want to tell me?

Swyx [00:07:27]: About mode collapse?

Diogo Almeida [00:07:28]: Oh, no. Oh, so mode clop-- collapse is related to this.

Swyx [00:07:31]: Yeah.

Diogo Almeida [00:07:31]: The disconnect happens because if you are in a mode covering or a calibrated distribution, you are, like, not. You are not overly punished about having outliers. You’d expect, like, something. 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?

Diogo Almeida [00:07:54]: Instead, GANs mode drop. They, like, drop the minority classes and just do the really common ones. And this is why this effect doesn’t happen, right? Like, instead of be-- in order to generate really long strings, without making errors, they need to, like, be extremely conservative because it’s e- 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.

Swyx [00:08:22]: Yeah.

Diogo Almeida [00:08:23]: And it’s, it’s a nuanced take and like, 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 are for decision-making is, like, a bad time.

Yann LeCun, JEPA, Scaling Laws, and Practical Research

Swyx [00:08:38]: And while we’re on the topic of Yann, do you agree that his fix i- with-- which is like a world model, like a JEPA-type, embedding thing is the right solve? So basically, like, the. One of the reasons that it could fail is because you’re trying to reason over token outputs and then, and then just looping back again and going. Keep, continuing going until you reach, like, a end of sentence. Like, is that, And his solve is JEPA, right?

Diogo Almeida [00:09:02]: Yes.

Swyx [00:09:02]: Which is, like, joint ambition,

Diogo Almeida [00:09:04]: Yeah

Swyx [00:09:04]: Joint embedding prediction. So like, is that the solve or, like, do you have a. Do you have a take on that?

Diogo Almeida [00:09:10]: 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.

Diogo Almeida [00:09:20]: Like, even my take here is practical. And like, I’m. Am I a scaling law fan? Depends. It dep-- it’s, it’s, it’s, like, it’s. Scaling laws tell you how much better you get at a thing for amount in.

Diogo Almeida [00:09:33]: A scaling law does mean exponentially more resources for normally sublinear gains, which looks to be a bad investment unless those, like, linear gains are, like, really valuable. But it’s all. To me, it’s all about, like, what can we do with what we have to make the biggest possible fucking difference? I can curse.

Swyx [00:09:51]: Yeah.

Diogo Almeida [00:09:51]: Yeah.

Swyx [00:09:52]: Yeah.

Diogo Almeida [00:09:52]: Yeah.

Swyx [00:09:53]: We’re, we’re, we’re approved for adults.

Diogo Almeida [00:09:54]: Hell yeah.

Swyx [00:09:55]: And also we have a scaling law thing if you wanna go into that later.

Diogo Almeida [00:09:58]: Oh, I could if we. See, that part is not super relevant right now.

Swyx [00:10:02]: Yeah.

Diogo Almeida [00:10:03]: I actually. If you wanna go into my bitterest lesson, I think that’s more relevant.

Swyx [00:10:06]: Okay.

Diogo Almeida [00:10:06]: But like, to me, I’m all about, like, pragmatics. And I think that the JEPA stuff is really cool early research. I really love awesome research. Is it practical yet?

Diogo Almeida [00:10:21]: Probably shouldn’t say. But like, there’s just a lot of.

Diogo Almeida [00:10:29]: I just think there’s just, like, so many diamonds in the rough let 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. Does it kickstart us as a company? Like, yes. Will it be great for us as a company? Yes. I think it’s gonna be, like, even greater for this direction of, like, programmatic AI. There was going to be, like, a gold rush on top of us for. ‘cause, like, software is super fucking charged. But I think there’s gonna 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?

Swyx [00:11:12]: Yeah.

Diogo Almeida [00:11:12]: And I think that’s why, like, the Twitter is just like, “Jev.”? It’s, it’s like. It is a party

Swyx [00:11:18]: It’s inspiring because it’s, it’s, like, 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?

Diogo Almeida [00:11:28]: That. Actually, if I. I’ll, I’ll make a tangent if that’s okay.

Swyx [00:11:32]: Yeah.

Diogo Almeida [00:11:32]: I think you might enjoy this.

Swyx [00:11:33]: Really? Our five tangents in. It’s good. It’s fun. Yeah.

Diogo Almeida [00:11:35]: Oh, yeah. I get lost at all my tangents.

Swyx [00:11:37]: This is gonna be horrible for the listeners to figure it out, but they’re gonna figure it out. It’s fine.

Safety Alignment, Refusals, and API Philosophy

Diogo Almeida [00:11:40]: Yeah, we can edit it in post.

Swyx [00:11:40]: This is my response. Yeah.

Diogo Almeida [00:11:41]: So popular thing on Discord, that people keep asking me, 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, like, a bot or whatever, you’re cloud coding, and a refusal happens, like, “I’m sorry, I can’t read DNA.py.” that’s an annoying time. It’s anno- it’s, it’s annoying

Diogo Almeida [00:12:18]: Right? But you can work with it, right? And you’re forced to work with it ‘cause of Stockholm syndrome.

Diogo Almeida [00:12:23]: 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?

Diogo Almeida [00:12:42]: Like, that is, like, straight-up insanity. It’s coming from a place of, like, people who do not understand software, do not understand programming, and like, they are obsessed with, like, I believe this, horseless carriage of, like, AI coworker instead of unearthing, like, the full power of AI.

Swyx [00:13:01]: Fair enough.

Diogo Almeida [00:13:01]: Yeah.

Swyx [00:13:01]: You want something that is the core kernel that is usable everywhere.

Diogo Almeida [00:13:05]: Yes. Exactly. Like, the cognitive core, right?

Swyx [00:13:07]: Yeah.

Diogo Almeida [00:13:08]: And you need this thing to be s- like, so general, so optimized for its use cases. You want it to be, like, you want it to work on all the future use cases, all the weird shit that people are doing.

Swyx [00:13:19]: Yeah.

Diogo Almeida [00:13:19]: We obviously didn’t train on any of that stuff. Is it surprising that it works? No, ‘cause we trained on weirder stuff, my friend.

Diogo Almeida [00:13:28]: So. But one tangent up about, like, safety alignment.

Swyx [00:13:32]: Okay.

Diogo Almeida [00:13:32]: Safety alignment makes sense for a product, in my opinion, for, like, ChatGPT and Claude. Like, it, What safety, what makes safety and capability alignment different is capability alignment is, like, 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 it and play around with it. Jeb is not anywhere close to that yet. It could be, but like, 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. But safety alignment is, like, the opposite of instruction following. It’s when you want to follow someone else’s instructions, like OpenAI and Anthropic

Swyx [00:14:13]: The RAGs value stack.

Diogo Almeida [00:14:14]: Exactly. And this makes a lot of sense for a product. Again, like, ChatGPT should do. Y- you sh- like, if they don’t want to, like, do, like, some, not-safe-for-work role play with ChatGPT, that’s on them because, like, maybe that’s, what their users who have, like, parents and kids want. Like, n- that’s fine. But in an API, that’s nuts, right? Like, that’s completely unacceptable because, like, people need to, like, program around this, and that is, that’s so anti-user that it’s. It. I’m. Huh. I can be an angry person, so I should try to calm down.

Swyx [00:14:52]: It’s, People get your passion, and I think that’s really good. The one pushback I’ll give you is, like, what if we use it to kill people, right? Like, that is the actual. Like, n- the not-safe-for-work thing, it’s private, personal, whatever. But like, yes, like, we will use it in war. And like, that is, something that companies can reasonably prefer their APIs not be used for.

Diogo Almeida [00:15:14]: 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.

Diogo Almeida [00:15:27]: 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. Will I do it at the technological layer? Absolutely not, because that will fracture the intelligence. Every single time you mean 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.

Swyx [00:15:54]: Yeah.

Diogo Almeida [00:15:54]: 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? Like, CIA. Actually, I don’t know what the CIA does, really. You can imagine. You can imagine, killing people who are not even bad or whatever.

Diogo Almeida [00:16:21]: 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 gonna 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

Swyx [00:16:46]: Yeah

Diogo Almeida [00:16:46]: 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.

Privacy, Benchmarking, and Trusting Intelligence

Swyx [00:17:11]: Yeah, that’s great. While we’re on the topic, let’s also briefly talk about your privacy stuff, terms of ser- terms of use, which, got a little bit of

Diogo Almeida [00:17:18]: Ooh

Swyx [00:17:18]: Misunderstanding. I just wanna clarify that upfront.

Diogo Almeida [00:17:21]: Hell yeah.

Swyx [00:17:21]: I think this probably takes two sentences from you about, like, you will not. You’re not being that restrictive about your API. Like, clearly

Diogo Almeida [00:17:27]: Oh, yeah. Oh, yeah, so yeah

Swyx [00:17:27]: Ideologically, you articulate your role as a platform very seriously.

Diogo Almeida [00:17:30]: Yes. Yes. I don’t know what you’re referring to, but like, this was. I’ve seen a couple of things about, like, benchmarking.

Swyx [00:17:38]: Yes.

Diogo Almeida [00:17:38]: Like, obviously we’re not stopping people from do. Oh, man, I should be careful about what I say. I’m realizing

Swyx [00:17:43]: No, you said, you said it publicly that

Diogo Almeida [00:17:44]: Yeah

Swyx [00:17:44]: That was in the preview period. You didn’t take it out for the launch.

Diogo Almeida [00:17:47]: Yeah. Okay.

Swyx [00:17:47]: And now you’re gonna take it out.

Diogo Almeida [00:17:48]: So the team is doing stuff that

Swyx [00:17:49]: Yes

Diogo Almeida [00:17:49]: 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.

Swyx [00:17:54]: Yeah.

Diogo Almeida [00:17:54]: 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. I’m extremely in fa- I’m medium about private benchmarks that are proxies. I

Swyx [00:18:09]: So are you worried about, saturation or, like, training on public benchmarks? So it’s, like, easy to cheat.

Diogo Almeida [00:18:15]: Not only is it easy to cheat, there’s a lot of ins. So I think that we are. Or anyone who’s, like, competition with us that, vaguely there is. Like, you could say, like

Swyx [00:18:28]: There’s like 50 Jev clones, yeah.

Diogo Almeida [00:18:30]: Well, sure.

Swyx [00:18:31]: Yeah.

Diogo Almeida [00:18:32]: Well, the, these. Let’s say that there is competition.

Swyx [00:18:34]: And we’ll talk about those. Yeah.

Diogo Almeida [00:18:34]: Or let’s just say that there’s. Let’s just assume that there’s an industry two 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 one. Like, people obsess about the cost and the speed. I believe 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’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 je ne sais quoi to it, right? Like, the good model smell. Like, the thing that happened after we launched of, like, two hours later that actually went way bigger than the video, which was like, “Holy shit.”

Swyx [00:19:16]: This is actually usable.

Diogo Almeida [00:19:17]: It. Well

Swyx [00:19:17]: Yeah.

Diogo Almeida [00:19:17]: It’s, like, beyond that.

Swyx [00:19:20]: Yeah.

Diogo Almeida [00:19:20]: Like, the. Whew, the launch was crazy, 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 je ne sais quoi, but the public benchmarks are extremely gameable. Even if they try not to, they still will. 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 with extra steps.

Diogo Almeida [00:19:58]: 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 ever-present part of o- 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. Like, if we wanted to, we could have released Jev, like, a year and a half ago if we wanted it to be dumb.

The Bitterest Lesson: Tasks, Data, and North Stars

Swyx [00:20:34]: Oh.

Diogo Almeida [00:20:34]: It. Like, the. My bitterest lesson, right? Like, architecture and Yeah.

Swyx [00:20:40]: I’ll bring it up

Diogo Almeida [00:20:40]: Hell yeah

Swyx [00:20:41]: Since you, since you talked about it, here.

Diogo Almeida [00:20:43]: Hell yeah. T- like, Sutton says that algorithms beats compute very roughly. Data matters way more than compute, obviously. And doing the right task, having the North Star is the hardest, most important thing. This has happened, in LLM land twice so far, right? Maybe 2.2 times. There’s RLHF, which, like, shifted the task to instruction following. No one realized 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, programs in the loop. And yeah, data matters so

Swyx [00:21:28]: Right

Diogo Almeida [00:21:28]: Unbelievably much.

Swyx [00:21:29]: So

Diogo Almeida [00:21:29]: Like, I can’t, I can’t emphasize it less.

Swyx [00:21:31]: Yeah, you consider yourself a data lab rather than, like, a model lab. Is that

Diogo Almeida [00:21:35]: Absolutely

Swyx [00:21:35]: Something. That’s the wording you guys use?

Diogo Almeida [00:21:37]: Yeah. We are. We will always, like, care so much about data. To me, model capabilities means data. Data is so unbelievably complicated, and that is what gets nines. Like, you have no idea how much data can shift everything. Data is so important.

TypeSafe as a Data Lab and Synthetic Data Strategy

Swyx [00:21:57]: Yeah.

Diogo Almeida [00:21:57]: Holy crap. So if people are looking for a job, we are hiring infinite data people, actually infinite.

Swyx [00:22:04]: What is a good data person? Like, clearly somebody who cares about reading through the transcripts of, whatever. You’ve said, for example, that y- all your data is synthetic.

Diogo Almeida [00:22:15]: Yep.

Swyx [00:22:15]: But that’s only, like, the scratching the surface, right?

Diogo Almeida [00:22:18]: Yeah.

Swyx [00:22:19]: Like, it’s not. Like, synthetic, so what, right? Synthetic, but we have people with a lot of taste and a lot of care looking at, looking at these, articulating what’s wrong, going back, regenerating. Is that what a good data person is these days?

Diogo Almeida [00:22:31]: Let me try to 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 ki. Well, I’ll do. Actually, number is zero. 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 environments, right?

Swyx [00:23:03]: Yes.

Diogo Almeida [00:23:04]: RLHF’s is the human feedback? Each task has its own unique kind of data, and we, of course, have our own unique kind of data, right? So number one, we have that. Number two, the thing I. 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 it will. We. 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, like, people, like, asking the same things where you’ll end up, like, overfitting to it and like, fracturing to it and all of that. And number two, we are, like, aiming for, like, a complete sci-fi future years from now where, like, these models are going to be, like, the general infrastructure, layers and layers and layers and deep down the stack to, like, things people can’t even imagine. Like, I would like to think of our model, like, 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. What we need is to, like.

Diogo Almeida [00:24:21]: It almost feels like a. Like, they’re the artists? They study this cognitive core. Our cognitive core is, like, 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, like, dimension, past, present, future.

Swyx [00:24:45]: The general case rather than the specific case.

Diogo Almeida [00:24:47]: Exactly. And like, that requires a lot of intelligence every time.

RLCD vs. RLHF: Defining a New Task

Swyx [00:24:50]: Okay, so we mentioned a little bit. You sort of criticized my thinking as r-- like, very RLVR influence, which is, like, very fair. Let us actually mention RLCD

Diogo Almeida [00:24:59]: Ooh

Swyx [00:24:59]: Which obviously you have some secret sauces to our knowledge. You’ve never actually published a paper or anything like that on it. No, right?

Diogo Almeida [00:25:05]: No, not yet.

Swyx [00:25:06]: But like, what should people get from this? Like, what. Can you give people some confidence that you’re just not just making up jargon for the sake of sounding cool, right? Like, one thing for me is, like, calibration I do think is a. To me, like, well understood because we’ve covered it in. On the podcast.

Diogo Almeida [00:25:22]: Yeah.

Swyx [00:25:22]: But I don’t know what you mean when you say RLCD versus what people are familiar with.

Diogo Almeida [00:25:26]: It’s a great question.

Swyx [00:25:27]: Yes.

Diogo Almeida [00:25:27]: And actually, I will give a related question.

Swyx [00:25:29]: Okay.

Diogo Almeida [00:25:29]: What is RLHF?

Swyx [00:25:31]: Okay.

Diogo Almeida [00:25:31]: Right? And actually, RLHF means multiple different things, right?

Swyx [00:25:34]: Okay.

Diogo Almeida [00:25:34]: Like, there’s the RLHF of the original. I think it was, like, Paul Christiano teaching a robot to backflip or something like that. Wasn’t there something

Swyx [00:25:42]: Was that it?

Diogo Almeida [00:25:43]: That was the original

Swyx [00:25:44]: I referenced the PPO paper, but I don’t know.

Diogo Almeida [00:25:46]: And so PPO was not necessarily from human feedback, if I recall.

Swyx [00:25:51]: Okay. That’s true

Diogo Almeida [00:25:52]: But I b- I believe it was, like, 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 teaching, doing PPO on language models.

Swyx [00:26:15]: This is the, sorry. I’m trying to, trying

Diogo Almeida [00:26:19]: Yeah

Swyx [00:26:19]: Trying to manipulate this thing. This is 2017.

Diogo Almeida [00:26:23]: Yeah.

Swyx [00:26:23]: Right.

Diogo Almeida [00:26:23]: I’m not 100% sure, but like, that looks quite right.

Swyx [00:26:26]: Yeah.

Diogo Almeida [00:26:26]: If it has, like, a robot doing backflips or something like that might be it. Yes. Okay, cool. I guess I got it right. Hell yeah.

Swyx [00:26:35]: There you go.

Diogo Almeida [00:26:36]: Yeah.

Swyx [00:26:36]: That’s the one.

Diogo Almeida [00:26:36]: So the idea was can, like, can you do, like, ill-specified things with it? So that’s, like, version one. Version two was, the learning to summarize work, that, like, OpenAI did, which is actually, like, PPO on language models to do something somewhat ill-specified. This is, like, another thing that people refer to as RLHF Which I did not co-author.

Diogo Almeida [00:26:57]: Oh, Dario’s there. Cool. Hell yeah.

Swyx [00:27:01]: And Radford.

Diogo Almeida [00:27:02]: Yeah. Shout-outs to Alec and Ryan. Love them.

Swyx [00:27:04]: Yeah.

Diogo Almeida [00:27:05]: But the thing that I refer to RLHF is the, Oh, man.

Diogo Almeida [00:27:13]: I’ll get to

Swyx [00:27:14]: You have comments on that, yeah.

Diogo Almeida [00:27:15]: I have comments on that paper, but like, we’re so many, tangents deep.

Swyx [00:27:18]: Yeah.

Diogo Almeida [00:27:18]: So the thing that really got. To me, the thing that I’m calling to RLHF is the task of instruction following. It’s not about the PPO. That part doesn’t matter. It’s about, like, setting a North Star of this is a valuable direction. It’s kind of like the Bitris lesson North Star.

Diogo Almeida [00:27:34]: And for us, RLCD is this new task. And it is not. I don’t see it as jargon. Like, I try to communicate with precision. It’s just that, “Hey, here’s another North Star.” Just like DPO and all of its, like, descendants also do RLHF, despite not using the algorithm in that paper.

Swyx [00:27:55]: And so clear- clearly stating the North Star is, being program- programmable AI is one, word that I really catch onto, removing the human in the loop,

Diogo Almeida [00:28:06]: Yes

Swyx [00:28:06]: From. Because RLHF is tuning

Diogo Almeida [00:28:09]: Yes

Swyx [00:28:09]: For this so that you can automate everything.

Diogo Almeida [00:28:11]: Yes. Everything that makes

Swyx [00:28:13]: Did I miss anything else in the, in the thesis of, like, what the North Star is?

Diogo Almeida [00:28:17]: There is. That is. That is right. I’m 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.

Diogo Almeida [00:28:30]: Right? Like, there could be programmatic types that are, like, sick AF, but if you. If the technology is not ready for it to. It’s not a tragedy if that’s not out in the world.

Why Programmable AI Matters

Swyx [00:28:41]: Yeah.

Diogo Almeida [00:28:42]: But to me, like, the pre-Jev world was a tragedy becau-- it sounds arrogant. Hear me out.

Swyx [00:28:49]: No. I strongly believe you.

Diogo Almeida [00:28:50]: Cool. It sounds arrogant, but like, I felt this way since long before I even had a company.

Swyx [00:28:54]: Yeah. I can, I can vouch that,

Diogo Almeida [00:28:56]: Yes, I’ve been talking about this for so long

Swyx [00:28:57]: You said this at All Around Her for, like, three years.

Diogo Almeida [00:28:58]: Yeah, I’ve been talking about this for so long. And I’ve been saying it because I thought it would have been easier. They say they do not do things because they. It. They’re easy. They. It’s ‘cause they thought it was easy, so

Swyx [00:29:08]: Yeah, exactly

Diogo Almeida [00:29:09]: Something like that. I thought it. This whole project would take a week.

Diogo Almeida [00:29:13]: And I was unbelievably wrong. So I am so sorry to everyone at OpenAI that I thought. I was like, “Man, I’m solving this right now.” but like, I think that the tragic thing is when. Well, I think overpromise, underdeliver is tragic too. And like, 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.

Diogo Almeida [00:29:40]: But like, it. To me, it’s like it’s just there’s just so much potential there. Like, AI is clearly so smart. I l- smart. I love this in my talks, when I ask people, like, “How can AI be so unbelievably smart? How can we, like, solve millennium prize problems in math, but still not automate even the most basics of works?” Like, really basic rote stuff that, like, the. It d- it doesn’t take, like, extremely smart people to do this. It’s not a satisfying job. Like, there’s other things these people could be doing, but yet we need them to do, like, this ba- like, super basic- non- unsatisfying stuff because, like, we can’t automate it yet, but we have this, like, supercharged engine of automation that just does not have, like, the right plugs and stuff to plug into all of this economically valuable work. And like, if the whole company of TypeSafe disappears, like, maybe it’ll take, like, a year or two for people to, like, truly catch up. I actually don’t know how long it’ll take. If model quality matters, then we are gonna be in a very good position for a long time. But it, like, it’s done, right? Like, there, like, this has changed the path of, like, technological history.

Swyx [00:30:49]: Yeah.

Diogo Almeida [00:30:49]: And like, we will be exploring that space as a field.

Swyx [00:30:53]: Yeah. I think, I definitely agree with that. You’ve created possibilities. So I think, if I can paraphrase so that people can un- also understand, you should not take the success of TypeSafe and Jev as just like, “Well, that is a new model type. Now we’re done. We go back to business.” Like, no. Like, actually, there’s, there are, like, five other model types that you should be exploring and like, let a thousand flowers bloom.

Diogo Almeida [00:31:15]: Absolutely.

Swyx [00:31:16]: Right?

Diogo Almeida [00:31:16]: Like, early internet

Swyx [00:31:17]: And some of that, some of which you will probably also build.

Diogo Almeida [00:31:18]: Of course, yes.

Swyx [00:31:19]: Yes.

Diogo Almeida [00:31:19]: Early internet energy. I think it’s back to tech utopia. It’s no longer like, “Oh, man, like, sometimes my coding agents work, but the, all of the best ones are hoarded internally.”

Swyx [00:31:29]: Yeah.

Diogo Almeida [00:31:30]: Right? It’s like creation is back on the menu.

Diogo Almeida [00:31:34]: ? Though it’s gonna be a wild-ass world, and buckle up.

Diogo Almeida [00:31:38]: It’s. And I’m so jazzed about that.

Manifesto, Launch Strategy, and Early Internet Energy

Swyx [00:31:42]: Yeah. And now you have the funding and the momentum to do whatever you envision there, which I, which I think is, like, very gratifying to see you have after, so long of saying these things

Diogo Almeida [00:31:53]: Yeah

Swyx [00:31:54]: But actually show the world.

Diogo Almeida [00:31:55]: I know. I just. Such a, such an interesting thing to be a tease the whole time. Like, my talk, like, felt like it was a cliffhanger ‘cause I didn’t say how the automation would occur.

Swyx [00:32:05]: Yeah.

Diogo Almeida [00:32:06]: Sean reviewed our manifesto And he’s like, “It’s a little bit vague in these parts.”

Diogo Almeida [00:32:12]: And like, “What’s step one? What is, what is the intelligence model?”

Swyx [00:32:16]: Well, I asked you for model, and you were like, “Yeah, model coming.”

Diogo Almeida [00:32:18]: Yeah.

Swyx [00:32:18]: And like, Well, I just, I mainly objected to the word composable But build prod.god is fantastic.

Diogo Almeida [00:32:24]: Thank you.

Swyx [00:32:24]: Yeah.