# Asking JEV's Creator the Hard Questions About This Hot New AI Model Markdown copy of https://realalpha.app/v/9Aato-NfjoU/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy. Diogo Almeida argues that useful automation needs AI optimized for reliable software decisions, not just human-readable text. - Show: The TWIML AI Podcast - Hosts: [Sam Charrington](https://realalpha.app/people/sam-charrington/) - Guest: [Diogo Almeida](https://realalpha.app/builders/diogo-almeida/) is co-founder and CEO of TypeSafe. - Published: Oct 6, 2026 - Length: 1 h 30 m ## The best minutes - [8:02](https://realalpha.app/v/9Aato-NfjoU/?t=482) [Choosing the right task matters more than adding compute](https://realalpha.app/s/choosing-the-right-task-matters-more-than-adding-compute/) (4 min) - [46:11](https://realalpha.app/v/9Aato-NfjoU/?t=2771) [Calibrated probabilities give developers control over AI decisions](https://realalpha.app/s/calibrated-probabilities-give-developers-control-over-ai/) (5 min) - [1:22:14](https://realalpha.app/v/9Aato-NfjoU/?t=4934) [Reliable workflows are a feature when automation must scale](https://realalpha.app/s/reliable-workflows-are-a-feature-when-automation-must-scale/) (5 min) ## Also worth watching - [23:56](https://realalpha.app/v/9Aato-NfjoU/?t=1436) [JEV's value comes from decision intelligence, not just speed](https://realalpha.app/s/jevs-value-comes-from-decision-intelligence-not-just-speed/) (8 min) - [1:17:54](https://realalpha.app/v/9Aato-NfjoU/?t=4674) [Agent designs can curate context instead of appending every step](https://realalpha.app/s/agent-designs-can-curate-context-instead-of-appending-every/) (4 min) - [15:47](https://realalpha.app/v/9Aato-NfjoU/?t=947) [Waymo's reliability comes from decomposed engineering](https://realalpha.app/s/waymos-reliability-comes-from-decomposed-engineering/) (5 min) - [54:04](https://realalpha.app/v/9Aato-NfjoU/?t=3244) [String-focused post-training can damage probability calibration](https://realalpha.app/s/string-focused-post-training-can-damage-probability/) (4 min) - [39:05](https://realalpha.app/v/9Aato-NfjoU/?t=2345) [JEV's training pipeline is built around explicit trade-offs](https://realalpha.app/s/jevs-training-pipeline-is-built-around-explicit-trade-offs/) (5 min) Skip the long generalization discussion, the model-versus-harness debate, and the closing product roadmap. ## Summary Almeida is co-founder and CEO of TypeSafe, which released Jev as a model for bringing intelligence into software. Before TypeSafe, he spent four and a half years at OpenAI, working on efforts including InstructGPT and RLHF, after earlier work at Google. Almeida argues that text-generating models are poorly suited to many software tasks because those tasks require structured, reliable decisions rather than human-readable strings. He describes Jev and RLCD as an effort to make calibrated intelligence available to software developers, with probabilities that let builders set their own policies. He says useful AI should be one component in an engineered system, not a single model asked to do everything. The discussion also considers how reliable decision models could change agent design and when businesses should choose engineered workflows over flexible but less reliable automation. ### Minute by minute - [1:53](https://realalpha.app/v/9Aato-NfjoU/?t=113) **Automation needs decision-shaped models.** Sam asks why agents struggle with basic data entry and insurance underwriting despite excelling at some research and coding tasks. Almeida says current systems center on transformers optimized to produce text, while accounting and similar work require discrete decisions meant for computers. - [8:02](https://realalpha.app/v/9Aato-NfjoU/?t=482) **The task directs ML progress.** Almeida reframes the bitter lesson: he argues that choosing the right task matters more than compute, and that data must be gathered for that task. He says RLHF required creating a new kind of human-feedback data from scratch, and that finding the right direction can lead to major advances. - [12:25](https://realalpha.app/v/9Aato-NfjoU/?t=745) **Demos still fall short of automation.** Almeida says he moved from Google to OpenAI, where he spent four and a half years on several projects, while remaining focused on why AI struggles with easy work. He argues that demos can look like science fiction without supporting dependable background use, and that some products fall back to human review. He points to Waymo as an engineering-led example and says ML should be scoped within a larger system. - [21:08](https://realalpha.app/v/9Aato-NfjoU/?t=1268) **Jev targets machine-native intelligence.** Almeida says Jev grew from his longstanding goal of optimizing intelligence for software automation rather than strings. He describes classifiers as useful interfaces for putting intelligence into software, not as an innovation in themselves, and says Jev aims to give developers flexible capabilities without assembling a specialist team. - [27:13](https://realalpha.app/v/9Aato-NfjoU/?t=1633) **Reliability matters more than cheap inference.** Almeida says speed and cost are not the core advance: developers are paying for intelligence, and a model only delivers what it was optimized to do. He says Jev's supporters recognized the value of structured outputs and programmatic use, and that TypeSafe has kept a high reliability bar even when it slows releases. - [32:19](https://realalpha.app/v/9Aato-NfjoU/?t=1939) **Building calibrated intelligence was hard.** Almeida says he initially thought the work would take a week, but producing robust, calibrated decisions required extensive work on model weaknesses and task design. He argues that developers need to operate at a useful semantic level rather than maintain many narrow cases, and describes Jev as a set of building blocks for intelligence applications. - [39:05](https://realalpha.app/v/9Aato-NfjoU/?t=2345) **Jev recombines existing model weights.** Almeida says Jev uses multiple open-weight models with different trade-offs and a process he considers more than conventional post-training; the field does not yet have a settled term for it. He says TypeSafe focuses on the intended task and removes elements tied to string generation, rather than pretraining from scratch, which he sees as resource-intensive for the gains. - [46:03](https://realalpha.app/v/9Aato-NfjoU/?t=2763) **Calibrated probabilities give builders control.** Almeida defines RLCD as a family of algorithms aimed at calibrated decisions, with probabilities that let developers apply their own policies in code. He argues builders should be able to set thresholds for actions such as refusing or escalating, rather than rely on inconsistent model choices or instructions in a system prompt. - [54:04](https://realalpha.app/v/9Aato-NfjoU/?t=3244) **Post-training can warp probabilities.** Almeida says the underlying models are more calibrated than RLHF and RLVR models, which he argues can distort probabilities in order to generate strings well. He says a calibrated-decision objective can build on models' existing capabilities to improve calibration and reliability. - [59:02](https://realalpha.app/v/9Aato-NfjoU/?t=3542) **Generalization depends on the objective.** Almeida questions whether decision tasks should be presumed harder to generalize than text, arguing that familiarity and the long tail of examples affect what appears out of distribution. He says ChatGPT already felt general for many tasks, while distinguishing generality from reliability on a particular workflow. - [1:03:05](https://realalpha.app/v/9Aato-NfjoU/?t=3785) **Training objectives shape generality.** Almeida argues that RLVR can intensify benchmark-shaped jaggedness, while RLHF was relatively easy to generalize because it optimized for subjective, human-pleasing responses. He says generality may be harder under RLCD and harder still under RLVR, while describing Jev as an early release that he expects to improve on business workflows. - [1:08:20](https://realalpha.app/v/9Aato-NfjoU/?t=4100) **Code should carry application logic.** Almeida speculates that better training data, rather than an agent loop alone, may explain improvements in coding agents, while noting he is reasoning from outside the labs. He argues that Jev should be called from users' code instead of relying on a universal harness, with builders supplying the application logic. - [1:16:25](https://realalpha.app/v/9Aato-NfjoU/?t=4585) **Agents could escape KV-cache constraints.** Almeida says TypeSafe wants to help agent builders make systems more reliable and efficient, though his team is not expert in identifying agent use cases. He argues that agents' append-only KV caches constrain routing, tool calls and flexible state, and imagines cheap, reliable intelligence enabling more selective context and control. - [1:22:20](https://realalpha.app/v/9Aato-NfjoU/?t=4940) **Rigid workflows suit repeat automation.** Almeida says one-off tasks may suit flexible, imperfect LLMs, but valuable background workflows justify engineered code and reliability. He sees being on rails as a feature for automation and expects hybrid systems that combine models with software workflows. For TypeSafe's next steps, he mentions additional models, early access and hiring. ## Transcript The TWIML AI Podcast, hosted by Sam Charrington, with Diogo Almeida. Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there. [0:06](https://realalpha.app/v/9Aato-NfjoU/?t=6) When it was released, developers split into two camps. Some shrugged and said it was just a classifier. Others saw something fast and cheap and easy to build with. Within days, open-source lookalikes were showing up, and even OpenAI has since announced its own response. But according to the person who built it, most of this discussion misses the point. Jev's real advance isn't speed or cost. It's a different idea of what AI should be optimized for. Diogo Almeida is co-founder and CEO of TypeSafe. Before that, he spent four and a half years at OpenAI, where he was part of the team behind InstructGPT and the RLHF work that helped turn language models into the assistants we use today. With Jev, he's now pursuing what he calls machine-native intelligence. [0:48](https://realalpha.app/v/9Aato-NfjoU/?t=48) Models optimized not for generating strings that people consume, but for making reliable decisions that software can act on. Here's Diogo on the gap he's trying to close. The elevator pitch I use for Jev is just where the f*** is all the automation. um i think ai is just so unbelievably smart yet so unbelievably useless at the kinds of things you'd really expect it to be useful for and most people as far as i can tell don't really have an answer on why there's like this entire bucket of stuff where ai is not just good it's like super humanly good like chat gpt deep research clod code but and on the other bucket of stuff where it's not just bad, it's like so bad that it can't even be used at all, despite having enormous financial incentive. I'm Sam Charrington, and this is the TwiML AI podcast. For over a decade, I've been exploring the ideas and innovation shaping the future of AI through conversations like this one that help you understand what's real, what's next, and what matters. Let's jump in. [1:57](https://realalpha.app/v/9Aato-NfjoU/?t=117) why can't people use agents for basic data entry and insurance underwriting? And actually, you know, if you go to, oh man, I don't want to be taking stabs at people, but like if you go even to Anthropics website and you try to use their agent, their customer support agent to do anything, it's really great at throwing the FAQ at you, but it is not useful if you want to do something even easier, like changing a credit card, you know? And that I think is a thing that everyone who is AI adjacent at least should have an answer to why that's the case. I'm kind of curious if you have an answer why there's like such a big divide between, you know, super smart stuff and like super, you know, useless stuff. I mean, I think part of it comes down to we've got this one tool in modern AI, right? It's this kind of token predicting transformer at the core with a bunch of machinery around it that makes it, you know, better at producing, you know, producing content, producing words, producing texts. [3:01](https://realalpha.app/v/9Aato-NfjoU/?t=181) And we're trying to kind of jam that into every place we can, but that's not always the best thing to forever. And I suspect that that's probably not too far from your viewpoint on this. A little bit, though. There's so much nuance into it, right? Like, for example, how can this token-spewing transformer solve millennium-price problems in math, but cannot automate accounting, right? There's a lot of nuance into that, where it really begs some questions that I don't think people in the field are really ready to answer and are very incentivized. At least the AI labs are very incentivized to not answer that truthfully is the implication there that you are ready to answer that question yeah i could if you want i mean like first of all like everything is just uh an opinion right so i i wouldn't say i have like proof of that i would say i have a lot of conviction and i have a lot of evidence but the way to prove this would be to actually just automate that job and i'm hoping that jev is like the first step in that direction so um my take on the answer and like the nuance of it is that you get what you optimize for which is just true of all things ml kind of all true of all things in the world things things all things yeah like [4:18](https://realalpha.app/v/9Aato-NfjoU/?t=258) actually i think ml people get it the least and people who interact with humans a lot get it the most um so you get what you optimize for and basically all the string llms have been optimized for strings and strings are meant to be consumed by humans or other llms but if you want something like accounting that's meant to be consumed by a computer, right? Like you have like specific fields and forms and discrete decisions. And it's very different to optimize for strings versus like these discrete decisions. And that I believe is the largest explainer of this gap. You know, when you kind of juxtapose Navier Stokes and accounting, that is like kind of a contrived example that kind of plays directly into the positioning that you've kind of built up around JEV and like what that's really good at. But I don't think that like AI being good at some things and bad at other things is like this one dimensional, you know, text versus decisions, like it's a jagged edge and there will need to be, I think, a lot of different things or something fundamentally different to smooth out that jagged edge and kind of push that frontier forward. And I'm wondering if you have any thoughts on what that looks like. [5:32](https://realalpha.app/v/9Aato-NfjoU/?t=332) can i actually test the hypothesis about the jagged edge for real absolutely because i feel like that's a fun one to talk about um as an aside a lot of what we do is really focus on making the model like this robust like cognitive core that can be plugged into any software and a lot of what we do is really focus on how can we smooth out the jaggedness that is put in via all sorts of like weird processes so uh but but like to test the hypothesis about the jagged edge existing um have you ever seen chat gpt respond to you rudely or like insulting you for no reason not if i didn't tell it to do that exactly so there's actually so the models are actually reliable at some stuff right they're not just they're like super duper reliable at some stuff i would say that jaggedness is not intrinsic to llms or ai uh you're saying jaggedness is relative to our application of them towards specific problems. [6:34](https://realalpha.app/v/9Aato-NfjoU/?t=394) Yes, and to our optimization process, right? Like if you optimize super hard in a particular direction, you get certain types of reliability that you lose others because you focused all of your optimization pressure on those things that really matter. Like if you were optimizing for decisions, for example, in a customer service chat, saying to give someone a refund when you shouldn't is from an accuracy perspective, exactly the same as like mashing your face on the keyboard you know it's it's the same zero percent accuracy but you never get them mashing your face on the keyboard because of the asymmetry in the optimization. [7:12](https://realalpha.app/v/9Aato-NfjoU/?t=432) This is what RLHF does, right? Like you use a reward model and the easier it is to punish something, the easier it will be to make sure that doesn't happen at inference time. And you just never get those types of things anymore, even though the pre-trained models could actually get word salad, right? So post-training can make certain types of things reliable. And I don't think people realize how much of an agency that AI practitioners have in setting their menu of desiderata into the model. It's super interesting to hear you say that. And you said that, you know, a few minutes ago as well, that everybody gets that you get what you optimize for, but ML people, but ML is all about optimization inherently. [7:54](https://realalpha.app/v/9Aato-NfjoU/?t=474) And, you know, all of our games with language models are, you know, RLHF. And like, you know, we talk about cost functions all the time. Like, how is it that you think that ML people don't get that? I love this. Have you seen my blog post on the bitterest lesson? No, tell me about it. I can just explain it. So there's Sutton's bitter lesson. I'm paraphrasing to be useful for me. So this is not exact, but roughly I'm simplifying. It's hard to simplify actually good writing, right? So I'm simplifying a bit to be beneficial to my narrative. Everybody else simplifies the bitter lesson. So you might as well too. [8:30](https://realalpha.app/v/9Aato-NfjoU/?t=510) Yeah, I might as well too, but I just want to emphasize that there's actually nuance to it. um but here it is so it's roughly speaking that compute and scaling matters more than algorithms and this is bitter because ml people love algorithms it's always the sexy thing to work on it's fun it's cool this is what they've studied um and it's bitter because you know like who wants to do that systemsy stuff and like getting gpus is hard and especially in the back in the academia days, you didn't have that much compute. And this birthed an entire era of scaling-pilled ML practitioners for mixed reasons. I actually think that there's pros and cons to being scaling-pilled. [9:11](https://realalpha.app/v/9Aato-NfjoU/?t=551) But I believe that data matters a lot more than compute. So you actually need data to put the [9:18](https://realalpha.app/v/9Aato-NfjoU/?t=558) **Diogo Almeida:** compute on. And more important than data is you need the right task. So this is the most important thing in all of the all of the ml that we do and the right task is kind of the north star that you [9:30](https://realalpha.app/v/9Aato-NfjoU/?t=570) get what you optimize for you know something like pre-training loss pre-training loss is like maybe obvious in hindsight it was quite a it was quite a big leap of faith that pre-training loss would lead to such awesome ai stuff and was by no means guaranteed so it's actually a rare few that i think can actually make a new task and that is the you get what you optimize for part of it And if I could add slight nuance to Sutton's take, in the case of games, there's no question what the right task is because that's constrained by the rules. And there's no data for games because you're doing on-policy RL such that you get data from compute by doing rollouts. [10:10](https://realalpha.app/v/9Aato-NfjoU/?t=610) So I would say that in the case of RL games, Sutton's bitter lesson is the equivalent one of what I'm describing. but by extension in real world uh application data is the constrained resource more so than compute figure out the direction of the data as well so um i won't talk out necessarily about jav and rlcd right now because you know it won't be fine we can but let's talk about rlhf right when we made there's many definitions of rlhf there was the original rlhf that paul cristiano did to teach robots to backflip. No LLMs involved. There was another RLHF learning to summarize where some people in the OpenAI team, I think Paul also was pivotal for that, taught the models to summarize, like a GPT-2 size model, to summarize text better than humans did. I'm not sure about that part. And then there's RLHF, which is what everyone refers to now. [11:10](https://realalpha.app/v/9Aato-NfjoU/?t=670) which is creating the task of instruction following, like teaching the models to chat and follow instructions. So when I refer to RLHF, this is a task that is being done. And before RLHF existed, there was literally no human feedback data on the internet of the shape that we were looking for because we had to create it, right? Like why would anyone have like preferences between two completions from language models that were somewhat on policy? Because that's such a weird thing to have. So it actually took some genius from the team to realize that this is a direction to go down. And once you have that direction, first off, when you're in the early days of the direction, you are not sure it's the right direction by definition, right? [11:51](https://realalpha.app/v/9Aato-NfjoU/?t=711) Like it's not yet working. You need like a lot of data compute and algorithms to make it work. And you just have a strong conviction that it's the way to go. But eventually you get the data and all the other stuff, but the data needs to be gotten from scratch. And that eventually becomes something phenomenal in the world. And that tends to be how you get like the gigantic jumps in ML. You know, chat to GPT was a gigantic jump. O1 was a gigantic jump. And Jev, at least, you know, I don't want to toot my own horn too much, but at least from like a excitement perspective was also a gigantic jump. [12:24](https://realalpha.app/v/9Aato-NfjoU/?t=744) Let's maybe take a step back because you mentioned RLHF. We spoke the last time 10 years ago, just a month under 10 years ago. It was October 2016. and we were talking about kind of deep learning broadly way before any of this Gen.A.I. stuff. Give us a brief summary of what you've been up to since then. You were at Open.A.I. for some period of time. I was at Google, and then I left. I was retired for a while, just trying to, like, chill out and be happy, did a lot of weird stuff, like very competitive video gaming and improv and stuff, and eventually I realized that actually AI is way more fun than that. [13:07](https://realalpha.app/v/9Aato-NfjoU/?t=787) let me join this weird ass company that has weird ass goals and that was open ai this is before language models were really a thing so i got the honor to actually work on a lot of open ai's greatest hits i was there for about four and a half years um and worked on a whole bunch of stuff but really even when i was working on other things like i've been obsessed with this question that has led to type safe and jeff like why is the ai so unbelievably good yet so useless for the easy work and that's that's been you know it's still an obsession if i'm like totally frank and it will not be done until we see massive massive transformative automation and software doing absolute sci-fi stuff there's so much so many things in there that i want to pick at like software is doing sci-fi stuff like clearly you're talking to your phone while you're great example on twitter guy is talking to grok and his tesla asking an agent to build a video like that's pretty sci-fi right software is doing sci-fi stuff and also the other thing to pick at in there is like the easy stuff like the easy stuff isn't necessarily easy uh which is why we haven't done it yet like oh man you can take whichever of those you want to take first and i [14:32](https://realalpha.app/v/9Aato-NfjoU/?t=872) don't believe in uh running from questions so i will try to get them both but actually i'll just do the first one i will ramble so i want to focus up on that one i think a lot of what ai does is super freaking sick i use coding agents i don't even code that much anymore but i use coding agents every day i use chatbots every day holy smokes would it be a lot harder to run a company without chatbots, but they are- And in case it wasn't obvious with the Tesla example, I didn't say the punchline, which was the car was driving itself at the time. Oh. Right? Yeah. Self-driving actually, I think it's a little bit different. [15:10](https://realalpha.app/v/9Aato-NfjoU/?t=910) front and we can talk about the distinction between self-driving was only self sci-fi like oh cool i i my um my argument with sci-fi is especially related to llms which i think tends to be what people mean by ai these days because it seems to be like the greatest compression of intelligence while having like the least utility so um and also it's quite i wouldn't say it's like universally hated but there's a lot of dislike for ai which is tragic and i think that we would be remiss as a field to not own up to why that is the case so there's like a lot of like dark sides to ai too and i could like name a subset of them but i think that that's worth talking about so number one i think self-driving is super duper cool um and also i see self-driving as a ginormous engineering win not necessarily an ai win um i use waymo as like the gold standard here no offense to tesla fans but like they seem to be safer and also the way they do it is by building like a la software and engineering like reliable systems that they understand the pieces of that have their own like decomposed abstracted parts and make sure those are like unbelievably good while programming in the behavior of the policies you want such that this can generalize [16:29](https://realalpha.app/v/9Aato-NfjoU/?t=989) outside of distribution so this is super freaking cool but none of this is happening in ai right now when in ai what you tend to get is a lot more let's just make a demo and you know the the my cynical loop here is make a demo raise like a seed or series a say that you're going to make it reliable never end up making that reliable pivot into a human in the loop version of this thing [16:54](https://realalpha.app/v/9Aato-NfjoU/?t=1014) **Diogo Almeida:** instead of actually automating the task and and it's not i don't hate on them to be clear that is actually the right move to do given the current ai climate because though models are not reliable enough so even you know i try to be as unbiased as i possibly can i obviously can't be perfect [17:13](https://realalpha.app/v/9Aato-NfjoU/?t=1033) But I would like to have like the same type of critical lens to LLM demos as JEV demos. And I, you know, like I think that a lot of them are super cool, especially when they have sick ideas. And also, I can't really vouch for them because they solve work when they actually reliably can be run in the background. And LLMs are extremely not there. I actually don't know of anyone who like actually runs like an LLM as a separate dependency because you can't get abstraction. with an LM, right? It can just break arbitrarily and you need to have a, the abstraction needs to leak in order to see like, oh, why did you talk about biology? [17:51](https://realalpha.app/v/9Aato-NfjoU/?t=1071) Oh, I'm sorry, my clod can't talk about biology. It fell back to another model or something like that. And that is extremely sensible for a first party app, but super nonsensical in my opinion for something developers are meant to build on top of. So I think that there's that, that lack of usefulness I still think is present there despite demos looking like sci-fi. because demos will always be significantly ahead of the curve. Opening I have seen has been showing like customer service demos since 2020. I believe that is now roughly six years ago and customer service is still not solved. And, you know, maybe because the work is hard, as you say, but it really looks a lot easier than Millennium Prize problems in math because I don't know anyone who could solve a Millennium Prize problem in math. [18:41](https://realalpha.app/v/9Aato-NfjoU/?t=1121) I feel like I know everyone, and I don't know everyone who could solve customer service, but everyone I know probably could do that job better than an LLM can. Or even drive-thrus. People keep trying and failing to automate drive-thrus, and there's no good answer for that other than maybe drive-thrus are harder than unsolved math. It's interesting to me that you call out Waymo versus Tesla in part because In the context of this conversation about like bitter, bitter lesson pilled, right? Because in one sense, like what differentiates Waymo is that they're not taking a bitter lesson pilled, you know, approach to it. Whereas Tesla's is more bitter lesson pilled. Like we're going to, you know, I'm characterizing their approach at a 30,000 foot view, but like we're going to collect a lot of data and throw that into a big model and let it do end to end. Whereas Tesla's, we're going to like classically engineer this thing, break this thing into a lot of different modules, use all the sensors we can to get as much data as we can. I've had conversations with Drago on the podcast about like how they approach it and how it's so fundamentally not like, you know, we're aiming for this end-to-end single model to rule the [19:55](https://realalpha.app/v/9Aato-NfjoU/?t=1195) entire driving experience. So I want to emphasize that that is just the ML part of it, but I don't think that's the limit of what we should be building. And I'm actually extremely pro software in software engineering. And I actually think that I hope that and I want to do everything I can that Jeb will usher in another golden era of software engineering, kind of like with the early internet creative energy of like building crazy things. And for ML, you need data, you need to focus on the right task, you need compute, you need algorithms. But ML is meant to be one part of a larger system, in my opinion, in general. And ML should be scoped to the part of the system that it can really do well. [20:35](https://realalpha.app/v/9Aato-NfjoU/?t=1235) And in general, the way to get higher reliability is to zoom in, you know? Like, would Waymo be as successful in San Francisco if they tried to do the whole thing end-to-end? Honestly, I would wager not. Like, maybe this is going to be better in five years, but in order to, like, actually ship something now that is useful and reliable and safe, I think you need engineering to study all of these things, study the properties of every single ML component, And when things break, be able to like debug which part caused it and fix that. And that's very hard to do in a one model rules them all type setting. [21:12](https://realalpha.app/v/9Aato-NfjoU/?t=1272) So where did the idea for Jev come from? Like, was this like, hey, I'm going to set up type safe and I'm running towards this thing that I already knew about. Did it evolve? Did you pivot into it? Like, how did how did it happen? Oh, I'll give you a fun story about this. Some investors, we've been fundraising and some investors have said, wow that's the greatest pitch we've ever heard and i told them that is i'm gonna avoid cursing bull bull that's bull um because i gave the exact same pitch last year and you guys didn't get it what you guys are reacting to is the product market fit so um i've been saying the same thing for years um which is what i'm now calling machine native intelligence like optimizing for automation which is optimizing for machines instead of strings and humans is exactly the same thing i've been talking about i did not know it would be in the shape of jev like we've had to get our own multiple breakthroughs to even get as far as we've had but the north star has always been what does AI look like if it's really, really optimized for being useful for software? [22:28](https://realalpha.app/v/9Aato-NfjoU/?t=1348) And JEV is our, you know, initial foray into that and our best guess of like the 80-20 of maximum usefulness. When JEV first came out, like, it seemed like the, it was a very kind of two disparate reactions to JEV and what it represented. it. You know, one camp of people was like, oh, wow, we've created classifiers. We've created logistic regression. You know, why is everyone excited about this? And there's another camp of people that was, you know, hey, this thing is amazing. Like, I could do things that I've been trying to do with LLMs for much, much, much cheaper, much, much, much faster. You know, part of my question is, like, what do you make of that? But, you know, more specifically, I'd love for you to talk about, you know, in kind of a technical level and with more nuance, like, how is JEV different from a classifier? How's it different from an embedded model plus a classifier? You know, OpenAI just came out with their decisioning model, which is like a small model, you know, post-trained for decisions. Like, how is it different from that? Like, what makes it? [23:38](https://realalpha.app/v/9Aato-NfjoU/?t=1418) and your approach unique oof um man i don't know where to start there because you've left so many breadcrumbs so i'm gonna just try to i'm just gonna jam eat them all i don't eat them all i'm gonna eat the most tasty ones okay so let me start with the tastiest one of them and i've had this rant a couple of times i don't know if i can do this rant without cursing so i'll try to control myself so classifiers not there's nothing wrong with classifiers classifiers are an interface like creating classifiers is not an ml innovation and no one in like 2010 to 20 to 2020 was inventing new shapes of classifiers they were just making classifiers for useful stuff um and classifiers are not like meant to be cool in ml they are literally the shape of usefulness yeah you know like that sounds like an arrogant thing to say but that is how you get yeah well Yes, that is how you put intelligence into software to make the software smart. [24:37](https://realalpha.app/v/9Aato-NfjoU/?t=1477) How is meta-powered? Classifiers. Maybe some regressors, but probably mostly classifiers. How is Google powered? Probably mostly classifiers. So there's nothing wrong with classifiers. I actually think if someone told us that we were a zero-shot general classifier for anything, that would be the greatest compliment I've ever been given for Jev. We have other things too, but for Jev, that would be an incredible description of it. Like, if someone could say that I can give you, you know, a random developer, the equivalent of an MLE team of a dozen people from 2019 to work on any task that you think up on the fly while coding, and it's instantly done, and not only that, it's going to be better than the quality in 2019, because the AI models have gotten a lot better. [25:25](https://realalpha.app/v/9Aato-NfjoU/?t=1525) it's going to be way way way smarter that sounds insane for development so i think that the classifier thing is um like grumpy ml people who kind of don't get it and that is totally fine because the jav party is a developer party oh man i should have said developer um so yes and not i actually think that there's another class of people who is not just like we can run the same stuff with lms so much cheaper i actually think that they are the they're the second wave of people i think the first wave of people who got really excited were the kind of people who felt like this is how ai should have been shaped all along this is like obviously the way to make stuff useful and i'm talking about like the people who like live and breathe structured outputs like the dspy i am not sure how it's pronounced dsp um but like i always say dspy but i think dspy is the most popular cool way of saying it but like these people have been kind of um you know trying to push this vision of this is how you actually make useful systems and i believe that they're intellectually quite correct but they've been hobbled by the models being optimized for strings instead of them being optimized for the thing that they want which is programmatic use so i think [26:45](https://realalpha.app/v/9Aato-NfjoU/?t=1605) those people not necessarily dspy but like there's a lot of developers who intuitively think in this way and part of what made us so popular is kind of on twitter them screaming like a phantom limb was returned and i you know i didn't expect this because i'm like this you know i i was designing for like where's my missing phantom limb and it turns out there's tons of other people with this And that was so... So freaking cool. That was only one crumb. Like, there are more crumbs out there. I don't even remember what the other crumbs were. Do you? I just got worked up. Yeah, embedding model plus classifier, small models, like... [27:23](https://realalpha.app/v/9Aato-NfjoU/?t=1643) The thing is, if you put a small model or an embedding model plus a classifier, a lot of people miss this. And my team has been telling me to stop telling this to people because it makes it more likely we have competition. And right now, I think everyone is missing this. i want to educate the world even if it's not in my financial best interest and the thing i really want to emphasize is that it is not about the interface the interface is cool it is not even about the speed and the cost speed and cost are bad things ideal speed is instant ideal cost is zero no one like spins up a library and says like i feel like putting cost and latency into this [28:00](https://realalpha.app/v/9Aato-NfjoU/?t=1680) **Diogo Almeida:** like line of code thing you want right so i think people just don't get that they're paying for intelligence and if you use a like a logistic regression on top of embeddings or a small model [28:12](https://realalpha.app/v/9Aato-NfjoU/?t=1692) you get the intelligence of logistic regression on top of embeddings or a small model right yeah but like everyone misses that you get what you optimize for of course you can get a gigantic jump in this thing if you're literally the only person doing that and we are literally the only person doing that despite me trying to shout out like hey everyone please do this this is really valuable for the world this is what's going to make ai really useful and you know there's been lots of copycats i still don't think people i was just going to say you say no one's doing this but like within the space of a week i think there are half a dozen or more open source projects uh a bunch of other like i mentioned at dev days yesterday open ai announced their decision models which is i think terra or a smaller model like that's about like the interface and the cost and the speed not about the intelligence right like if people really want to taking the same approach that you're taking to get there they don't even need to have the same approach they just need to care about intelligence you know like if you're taking like an open source model doing something to make it have the same interface and same characteristics and you just benchmark on nlp [29:22](https://realalpha.app/v/9Aato-NfjoU/?t=1762) tasks or something like that like you don't think i thought of that this was like literally day one into this project uh it's like i i i thought this project would take a week um it's been fun you know i i only feel okay admitting it now totally honestly because it's been so successful like i it becomes feels way safer to admit my feelings at the time we started this i was probably at worst tied at best at post training in the world when we started and i thought it have taken a week because i've post-trained so many models for so many different things and models are they seem to be very very general at a lot of stuff but apparently not decision making which i've learned the hard way so i i i think it's very cool that people are playing around with stuff i worry that if people play with bad versions of the model they might actually get burned by the whole subgenre and say like hey like like this kind of stuff is crap i i hope that's not as big a deal as people were saying that kind of exactly because the thing they want is intelligence if the models were perfectly intelligent which our models are not and no one else's are then you could literally solve everything classification is ai complete right [30:37](https://realalpha.app/v/9Aato-NfjoU/?t=1837) like any task could be boiled down into classification questions and you could just solve everything so truly intelligence matters in what you can and can't automate and what kind of sci-fi you can build and it's it's really important it's so important i try so hard you know like we could have released so much earlier if we didn't care about this and you know that even now it's a debate for like shipping new stuff how do we continue to ship stuff when we have such a high bar for reliability because we want to be trustworthy we want to be trustworthy to developers not ml researchers who are like you know snoot about which benchmark things are better at. We just want people to not have to worry about their intelligence because that is what developers want in order to build stuff. So I get worked up about this basically every day. It's such a frustration. And people on the street recognize me and they're like, oh, you're the Jeff guy. I love how smart and cheap it is. And I'm like, no, what has my reputation become you know like i want to be a paragon of making ai so reliable that it's boring like sequel you know like i want ai to be so predictable that you can like write queries without having to even run them against like eval sets because you know this is what common [31:54](https://realalpha.app/v/9Aato-NfjoU/?t=1914) sense intelligence would do there and to just do it every single time and that is when software engineers get super duper powers and yeah i i obviously love jev i love that we're fast and cheap we're gonna keep pushing that just you wait and also i cannot emphasize enough reliability is what people are paying for reliability is what people want um and if you want to put intelligence to applications you need the intelligence so let's let's dig into how you get that intelligence um Um, you know, I think maybe let's like start at the end state, like what it is, how it works. Um, you know, we can from there maybe touch on RLCD, which is a big part of your process. [32:36](https://realalpha.app/v/9Aato-NfjoU/?t=1956) Uh, but along the way, I'd love to hear, like you thought it would take a week. It took a lot longer, how much longer, and maybe what some of those, some of those milestones and turning points were. Oh man. And on that note, if you want like a fun, embarrassing thing about me is at the time I had this idea and like the aha hit me, I wrote a document in OpenAI entitled, titled, maybe not entitled, titled, we could have had AGI last year. and it was meant to be very spicy where it's like and actually i did believe it at the time i still believe the sentiment to it but like the the scent the sentiment inside of it was like one year's worth of time of like the equivalent of an rlhf effort would be sufficient to make this new class of models that are made to be machine native and um rlhf could have been done in less than a year for sure um in hindsight we we've learned but um in hindsight i think rlcd could not have been done in a year it was just much much harder i think that making like lms are much more suited to to making things like human pleasing than ai is to be making calibrated robust decisions and there's so much jaggedness when you remove the psychophantic elements of ai that it's there it's it's been a lot of work is what i would say and i see the north [34:09](https://realalpha.app/v/9Aato-NfjoU/?t=2049) star um and as and of how we do things is really really really trying to find what are the ways like if you imagine the cognitive core that has like these jagged cracks what is the most efficient way to like heal that to get smooth robust performance at whatever level of intelligence is truly capable within the cognitive core. So it's a lot of like understanding the capabilities of LLMs. It's a lot about like seeing their weaknesses. It's a lot about like figuring out what weaknesses are addressable and how do we address those? Our model cannot count R's in strawberry. Luckily, that is not what real world work looks like, but it can do a whole bunch of other stuff because a lot of work is in the shape of semantic judgments, which luckily LLM... [34:52](https://realalpha.app/v/9Aato-NfjoU/?t=2092) You can't decompose that into a classification problem? Oh yeah, you can decompose it, but you can't just ask it vanilla. and at some point like it's a little bit i'm just teasing yeah let's not put it's cheating right like if you like break it into like a list then it works um so that the tokens don't get merged but like if you you want like a sane level of decomposition that makes sense for your engineering system that's ideally at the right level of abstraction that it's understandable by an engineer and it's maintainable versus like overfit to any one question so um even if our model could solve every single task if decomposed perfectly that may not be sufficient because we want to keep moving up the abstraction hierarchy so people can program at the semantic layer that matches the way they think about it, right? [35:42](https://realalpha.app/v/9Aato-NfjoU/?t=2142) Like if you could break down, like does this person want a refund into a hundred different refund scenarios, like maybe then that'll work. But like that's still kind of a pain in the ass where some businesses will still do it. But like that, I wouldn't consider our job done for that. We've talked broadly about Jeff and it may be that everyone who's listening to this has already seen it, played with it, knows a bit about it. But in case not, you know, So I asked you a bunch of questions before. Let's kind of break those down. The first is like, from the perspective of a developer, your target user, like, what is this thing look and feel like? [36:20](https://realalpha.app/v/9Aato-NfjoU/?t=2180) Like, I think of it broadly as like, or you could call it an intelligence API. You give it a bunch of possible decisions and some input, and it like makes one of those decisions or tells you which I think it is, and it gives you some probabilities. How would you tune that? I love it. A lot of people describe it in many different ways, and I take all of them as long as they're accurate. I also like the description of SQL for intelligence, you know, or like it's a combination, right? Like system, there's the interface and the model, and we are kind of made both, so it's a little bit confusing. But to me, I see the interface as like the building blocks for making intelligence applications, kind of like adding a fourth logic gate or something like that. [37:04](https://realalpha.app/v/9Aato-NfjoU/?t=2224) so i really want intelligence to be standardized into a common set of primitives and ideally everyone makes these things as smart as possible not just us i also like the kind of positioning as like hey we have this thing called language models here's this new thing called decision models um that kind of says a bit about uh what it uh aims to do not necessarily jev but like some emerging class of yeah tell me i will get a tiny warning about this where we consider decision models as a name i'm not opposed to the name i think that the best categories are named by the community so um but i will give a disclaimer that there's a reason we didn't pick that and i think that there's a lot more that fits this class of models beyond the decisions so i would say that like people's public interface to jev could be called decision models but i wouldn't yeah i personally wouldn't call what's an example of something that jev it would be good for that's not a decision oh i will not tell you but i will give you a hint if you'd like um so um i've said that decisions and classification is how you make a intelligence useful to software and we are all about making intelligence useful to software um and i'm also saying that there's more than just [38:30](https://realalpha.app/v/9Aato-NfjoU/?t=2310) decisions available which sounds like a little bit of a paradox but i will say it will one day be obvious um and i can't reveal that yet because these are like things like a little bit more distant in the roadmap. But it's a fun one to think about if your readers want to noodle on this. So that's kind of the shape and kind of functional surface of it. Let's talk a little bit more about kind of the making of, like at the core, there's a model that we won't call a decision model, we'll just call it JEV. And it was created through a process called RLCD post-training. Is it a post-train of something else? [39:13](https://realalpha.app/v/9Aato-NfjoU/?t=2353) Is it a ground-up thing? Like, what can you tell us about this thing? I will answer this in a way that is, like, overly precise. Back before RLHF existed, post-training as a concept did not exist, right? Like, we as a team, we didn't name post-training, but, like, that we created the concept of post-training. At the time, fine-tuning existed, and people were like, are you just fine-tuning the model on strings? And by some definitions, we were fine-tuning the model, but we did a whole lot more, such that at some point it becomes misleading. so i would actually say that um just turning my oh there we go um i would say that calling it post-training is probably going to be seen as inaccurate in hindsight but the field doesn't have a terminology yet for what we are doing so uh that will probably be revealed more in the if you were to invent one right now what would you call it no one likes my ideas in the team um so i've said this elsewhere so i'm okay saying this but we okay back to bitter's lesson we because this is ml we care a lot about the north star and we are willing to do everything for it and i try to be as upfront about the trade-offs as possible and part of that trade-off is being [40:37](https://realalpha.app/v/9Aato-NfjoU/?t=2437) a lot worse at string generation and even like the jev can chat stuff like it's very cute but it's obviously dumb uh like these are trade-offs we intentionally make so when designing the entire pipeline um post training was a lot easier on top of llms because the shape was very similar like it's strings here it's basically strings there technically there's some structure with like the chat ml format but it's mostly just strings you know so there's a lot of similarity in the stack what we're doing is we're actually like ripping out a lot of fundamental pieces and like rejiggering them and using pieces from other stuff so i have described it as a a frankenstein's monster of models um which i i haven't read the book but i've been told is at least innocent if not the good guy just to be clear because people are like don't call the model a monster and i'm like it's a good analogy though what's the foundational thing that you're applying this unnamed thing to is it like like an off-the-shelf Quen model or something? [41:39](https://realalpha.app/v/9Aato-NfjoU/?t=2499) A lot of different open source models because they all have different trade-offs to them. So we are actually Frankensteining all over. But the important bit, if we want to be maximally precise and useful, [41:50](https://realalpha.app/v/9Aato-NfjoU/?t=2510) **Diogo Almeida:** is figure out the North Star, figure out what you're willing to trade off and remove all the stuff that is useless for that trade-off [41:57](https://realalpha.app/v/9Aato-NfjoU/?t=2517) or that is needed for the thing you're giving up and hyper-focus on the thing that matters. And a lot of it will follow a la first principles with months of research sure but that's that's broad in general and to be concrete and specific it is interesting that like you didn't start from nothing like you started from llms and like transformers and like you're building on these building blocks that have established to be clear um i have said you did or did not i did not say transformers i did say open weight models slight distinction um which i would just want to be precise um so i think that oh man something i want to event well number one it's very obvious we haven't done pre-training like if we could pre-train better than the rest of the world with the budget we've had then um Um, the rest of the world is totally cooked. [42:56](https://realalpha.app/v/9Aato-NfjoU/?t=2576) Um, like we, like, uh, I, I, I think free trading is not the most efficient thing in the world. And I, it's, it's, you know, exponentially more resources for. debatably linear gains seems very sublinear um it's kind of a bad deal in my opinion unless it's necessary um meaning we've kind of already boiled the ocean of content down into things that are kind of consumable by some downstream process why keep doing it unless you're doing something fundamentally different exactly so we are trying to focus more on like the 10x or more jumps that can be made. And we think that that can be very much done in a budget. Well, we obviously think that because we did. But I think that all AI companies should be held to that same bar personally. Part of the reason why I'm trying to really drill down on this point is not necessarily for you to say, oh, we use, you know, Quen 3.6 or whatever it was, is that it took a while for me to wrap my head around the fact that there was an LLM at the core of this thing and that is the source of intelligence and that is the source of zero shot and like if you think about it as like like if all you've seen is twitter and like oh it's like this cool new classifier thing and you're like oh okay well that could be like i don't know xgboost or like [44:25](https://realalpha.app/v/9Aato-NfjoU/?t=2665) some other it could be anything and like they've done some like really training thing but then you you are like oh well then i have to like post train it or i have to like give it some information about my domain or something like that how am i just going to give it a bunch of decisions and ask it to decide like where is that intelligence coming coming from to your point about intelligence uh but then you can start to see how that could work if you realize there's an llm down there somewhere i i could see that analogy like i i kind of um i i'm trying not to to avoid the llm terminology because it's technically not a language model you know like it doesn't model language despite people wanting to force it into that i see i kind of see this as like this more broad like this recent wave of ai has been creating these incredibly awesome condensed bits of intelligence of which, you know, like with a scaffolding around it that turns it into an LLM. [45:29](https://realalpha.app/v/9Aato-NfjoU/?t=2729) So imagine like this mega battery with like stuff on top that like turns into a language model. I only meant it in the sense of like a compressed version of the internet. Exactly, exactly. So I would use that. You've got a compressed version of the internet that some large effort created and you're using that. Exactly. And I think that's the only way to get what I'm calling general intelligence. That's the only way I've seen to get to general intelligence so far. Or if you, you know, it's debatable if you want to call it general intelligence, but I think it's general. Not to oversimplify, like you've done a lot of things, like you've Frankenstein, it sounds like several different models. [46:08](https://realalpha.app/v/9Aato-NfjoU/?t=2768) You have an involved process that we won't call post-training, but there is this thing called RLCD. What is that? So back to the topic of RLHF, like RLHF is not about PPO on reward model on, you know, like doing first SFT and then reward model training and then on policy training of that. Okay. RLHF is about human decisions and your thing is about decisions, calibrated decisions. Human preference. It's never about human decisions. Sorry. Yeah, yeah, yeah. Human feedback. back. So like when I talk about RLCD, I talk about like the general class of algorithms that optimize specifically for calibrated decisions, because that is what's useful for software. [46:51](https://realalpha.app/v/9Aato-NfjoU/?t=2811) So we do have many algorithms in here that we have been playing around with. We actually have talked about publishing something here, though I will not promise anytime soon because the team is super duper swamped. But I think that you get out, you can imagine. But I think that We'll be right back. So, publishing RLHF was probably really good for the world. You know, it allowed a lot of people to understand stuff. It took us a couple of months to publish it. But, you know, we had InstructureGPT for a while. And I think that eventually publishing something here would be really good for the world too. So, I think that that could be in the cards. [47:30](https://realalpha.app/v/9Aato-NfjoU/?t=2850) But the task, that's the algorithm versus the task, just to be clear. Other people will probably make other algorithms within the same umbrella of the task. Is it useful to talk more about the task or do you mean the task broadly in the way we've been talking about it thus far? Like what, like in an academic context, I'd ask about the setting. Like what is the setting for RLCD? Like how have you defined that? That is very complicated, but I can give a simplified version of it. Have you ever used like function calling in the chat completions API? So function calling is kind of like decision making because it is decision making really, but it's done on a layer above the string stuff. [48:20](https://realalpha.app/v/9Aato-NfjoU/?t=2900) So all the knobs available are the stringy knobs that are still available from like the string models. And one example of a stringy knob is a logit bias. A logit bias allows you to like nudge in the direction of sampling a particular token. So you can do that to make certain tokens less or more likely, if that makes sense. And one of my claims is that no function calling API should exist without a logit bias for the function itself. Like, why would you not provide that knob to a user? because if you are, for example, doing a customer service thing and you need to figure out if you're escalating to a human, your policy is probably extremely different if you're Walmart versus like Costco. [49:06](https://realalpha.app/v/9Aato-NfjoU/?t=2946) I know nothing about these policies, but I'm just making some of them. But like the only alternative now then is to like program again into the language, like please give refunds or please don't give refunds or something like insane instead of giving the controls of the system to the programmer [49:23](https://realalpha.app/v/9Aato-NfjoU/?t=2963) **Diogo Almeida:** implementing the downstream behavior so to me rlcd is about exposing the right interface to builders so that they can get the properties they want without just praying to the system message [49:35](https://realalpha.app/v/9Aato-NfjoU/?t=2975) and like i like these types of basic things are table stakes like uh this hasn't happened recently but this happened back in the older days when open ai had issues with over refusals they had to like roll back the model because they refuse too much. And I would argue that that was actually bad decision-making. What was happening is the model within its weights was making a decision, refuse or no refuse. And then the system is just going with what the language model thinks because it blends the decision with the text and then just does that. So there does exist queries by construction that if you refresh it several times, it will sometimes refuse and sometimes not refuse that is insane behavior from like from an engineering point of view right why wouldn't you have a threshold why wouldn't you have a tunable threshold why would you not have a tunable threshold that responded at least somewhat indirectly to how people are complaining right or if you're doing customer service why can't your threshold on whether to escalate or not be dependent on really basic things like how much is this customer worth to you to us and how busy are our, you know, human workers right now, because maybe that changes things. And we just [50:49](https://realalpha.app/v/9Aato-NfjoU/?t=3049) don't have any of those controls today. So I see RLCD as really exposing the most native to engineering controls. And that is the shape of the task. And a lot of stuff comes from that shape. Like in order to have meaningful knobs to tune and thresholds to chop, you need to be calibrated. you need your probabilities to be like smart and accurate. You care about the long tail of probabilities, not just the accuracy. And there's like a lot to the task itself on studying this, seeing trade-offs, figuring out what the right behavior is. There's a lot to the product of an ML API that most people don't realize because they mindlessly follow the gradient of things like MAUs, monthly active users. [51:36](https://realalpha.app/v/9Aato-NfjoU/?t=3096) You talked about the RLCD setting being one of you've got some kind of some set of inputs and and, you know, there are the inputs that the user provides, but there's also these knobs that kind of control internal bias in the system. does the current jev product expose those to the user or are those knobs that the jev engineering team has access to to create this thing that uh you know this decision api that you've published to the world as far as i can tell the internal api and the external api are basically the same so we want we but how would we know what users should set this to if we don't have their application right like what is the right confidence is part of this probabilities are part of this most people don't use the probabilities but i would guess when things get into hardcore engineering you would definitely use the probabilities every time over just picking what the model thought was best because even if all of the choice even if you had like the perfect probabilities for each of the actions you can take as a company or a business deploying this or even a software engineer like trusting it with your own home system, each of these actions have different costs and benefits? [52:54](https://realalpha.app/v/9Aato-NfjoU/?t=3174) Well, I guess I want to come back to the probabilities as outputs. I'm more thinking about like input knobs that when I've, you know, my cursory look into the JEV API, it's like a very simple API shape, but I've not dug into the docs and looked at all the possible knobs that you give me access to beyond just the decisions that I send in? Like, what are some of those knobs? The knobs are via outputting the probabilities. We are basically trusting the user to make those decisions. This would be equivalent to, imagine speed wasn't an issue. It would be the equivalent to the OpenAI API outputting logits for every token. [53:36](https://realalpha.app/v/9Aato-NfjoU/?t=3216) And then the software developers can choose how they want to chain the sampling algorithm instead of letting OpenAI just handle everything. So we give those knobs by giving people a raw access to the models than other people have access to the string models. And that way they can tune their behavior to what actually is needed in their system instead of what, you know, like OpenAI PM thinks is like a globally okay behavior for everyone using the API or first party product. Which brings me to a question I had about how you do that. and i guess i imagine this is circling back on rlcd um but like historically uncertainty estimation has been like very difficult um and you know particularly when we are talking about these you know large compressed versions of the internet that historically generate text and like uh and large complex models generally like so how do you how do you do that hard work blood sweat and tears and also and also like you know it's hard work like we were in stealth for a long time you know being two years in stealth is like kind of insane in in like ai terms the amount of i'm not asking you to divulge the family jewels it's more like is it is it grounding on something like [55:03](https://realalpha.app/v/9Aato-NfjoU/?t=3303) What's the guiding star, the like the principle that kind of helps us? Understand why we should trust these probabilities as opposed to, well, they've just got some token generator that's been trained to generate numbers between zero and one. I'm going to give you like one more bone in the last thing, which is, you know, we did work in stealth for a long time. We missed so many fads, even though we thought Jev would have been great for a lot of those fads. Like, holy shit, it would have been cool to launch Jev at the same time as OpenClaw. It would have been awesome because everyone has like a lot of reliability issues And also, these compressed things of the internet are, as far as we can measure, decently calibrated. [55:46](https://realalpha.app/v/9Aato-NfjoU/?t=3346) They're not perfectly calibrated, but they are way, way more calibrated than RLHF models are. [55:52](https://realalpha.app/v/9Aato-NfjoU/?t=3352) **Diogo Almeida:** Actually, RLHF and RLVR destroy the calibration of the models immensely, because in order to output strings well, you actually need extreme overconfidence. [56:02](https://realalpha.app/v/9Aato-NfjoU/?t=3362) It requires a mode-dropping loss in order to make that work, which just completely warps the probabilities and i think that the gpt4 paper started talking about this because that's when models started not shipping their pre-trained versions and only started shipping post-trained versions so i think the gpt4 paper went into it a bit but the models used to be decently calibrated look at a lot of opening um anthropics early publishing you know measuring uh with what they called pino the probability it knows facts like these are all like pretty decent and it's definitely not good enough i would guess jeb would crush all of those like super easily the the thing is that there's ways to bootstrap calibration and reliability and intelligence from all of that the same way that rlvr bootstraps reasoning right like in an ideal world the reasoning of rlvr doesn't come from humans writing the reasoning because it does it differently but if you have the right rl objective of calibrated decisions there are ways to get that working from that seed data into becoming more and more and more calibrated over like the training that i don't really know what it's called epochs training stages whatever so there are there's like hand wavy ways that makes sense [57:22](https://realalpha.app/v/9Aato-NfjoU/?t=3442) that you can get calibrated stuff out of not quite nothing but out of like the beautiful raw material that AI has today, because AI has like a lot, like, I feel like AI is still a diamond in the rough. You know, there's so much stuff to do there. And if I wasn't doing GEV, I would probably be ideally not starting a company, but doing other cool stuff on top of the AI models. Like, how do you get to generalization? Like, you know, presumably you've got, you know, you're grounding on some corpus of decisions so you can teach this thing like what decisions are like and, you know, how to map to these probabilities and all this kind of stuff. [58:01](https://realalpha.app/v/9Aato-NfjoU/?t=3481) But like that, you know, as with all these training efforts, like, you know, you have to be asking things that are in distribution in order for the results of the model to be representative. And like, how should we think about, you know, if you're trying to enable, you know, zero shot, you know, classification, like zero-shot decision-making, how do I know that, like, my particular problem is in distribution versus out? And maybe more broadly, like, I would think that, like, generalization would be a harder problem for workflows and decisions and, like, software interactions and all these things that, like, we're trying to automate than uh than like just generating text but like i could be way off on this so you tell me it's an interesting question which maybe would be more fun as a dialogue than a lecture um you know like maybe like why do you think AI lacks generalization? [59:12](https://realalpha.app/v/9Aato-NfjoU/?t=3552) I have my own answers that I could say later too, but what makes you think that a generalization is not native to AI? It sounds like you're asking a philosophical question. I can ask a practical version. How general did ChatGPT feel? I think the perspective that the question came from was one of, as a general principle, Like the reliability of responses is governed by kind of mapping the training data and like, you know, the degree to which the thing you're asking is familiar to the model, you know, using these terms very broadly and like very broadly. right and so like i think why the take that um you know decisions would be harder to generalize than uh than text is because like presumably i get enough texts in the world if i'm asking a model to generate texts like it starts to understand like how to put words together and if I'm not asking it for precise facts then you know it can like put words together that are plausible whereas like in the case of you know I've got an accounting workflow and I needed to do a very precise thing like if that workflow is very different from the standard way of doing that workflow like i would think the model would be very confident and wrong oh so fun fun stuff um [1:01:03](https://realalpha.app/v/9Aato-NfjoU/?t=3663) so you know like it was chat gpt general is one one way i would like to start the conversation did it feel general to you i mean i think it i think it is general like and i mean this is maybe going in a you know a different philosophical direction but like you know i was very very strongly against agi as like a a broad term for like what we've achieved but like in the past three months maybe like i'm getting closer to being able to say that like we've achieved this like in the sense of um you know i can talk to an agent that you know you know conversationally and have it do things in my business like an employee might. [1:01:48](https://realalpha.app/v/9Aato-NfjoU/?t=3708) Like, it is like AGI and generalization are different concepts, but like I'm accepting of the idea of ChatGPT being general. Yes. I don't know that that like negates the in-distribution, out-of-distribution line of questioning. I mean, there's a bit of it because in-and-out-of-distribution is also a function of like how long and hard you've worked on that because you start mining the long tail of examples so in the grand scheme of things chat gpt of today is a lot even if you use instant mode so turn off reasoning is a lot more um refined than chat gpt of four years ago well three years and 10 months well sure but not necessarily because the model is more general but because now we've built these harnesses that know how to go out and find information to fill in context. [1:02:46](https://realalpha.app/v/9Aato-NfjoU/?t=3766) I mean, if you talk about just ChatGPT, I would imagine that there actually was generalization gains, but ChatGPT already was quite general and it did a lot of simple stuff quite well, I would argue. And I do think that there's some amount of generality native to things. I actually have a separate blog post. I've called, I believe it's called Lies, Damned Lies and Benchmarks. Where I talk about how a lot of the fragility and jaggedness we see of LMs today come from RLVR. Because if you are being non-generous to RLVR, it is kind of, by definition, optimizing for benchmarks. like by definition a verifiable reward is a benchmarkable one and by adding latent reasoning before the answers you give the models all the power in the world to do really well in those benchmarks so it's very unsurprising it does very well in benchmarks i think a lot of the jaggedness has come from this and and then you get to this whole like uh like challenge of like adding more and more and more and more and more data in distribution to get more and more diminishing returns to make the jagged frontier even more fractal and spiky such that it can do some things really reliably but not super duper and i i think it's gonna be really interesting [1:04:09](https://realalpha.app/v/9Aato-NfjoU/?t=3849) to see how this plays out because they're spending like a lot of money on this and it seems like it's not yet super general is what i would get the sense of and i think that our path to generality is going to be a lot easier from what I've seen. But by far the easiest for generality was RLHF. RLHF was ridiculously easy to make it general because it was very, very good at writing human-pleasing stuff because that is subjective. So I think that this is one of those, you get what you optimize for, but you need to be aware of the real world. And the real world is how good are like these compressed versions of the internet at surfacing intelligence. [1:04:48](https://realalpha.app/v/9Aato-NfjoU/?t=3888) and i think that they're incredibly good at writing text to please people they are quite good at system one tasks but they are it's it's quite a battle to get them good at the system two tasks hence the extreme jaggedness that we're seeing right now it'll be really interesting to see where it goes though i'm very excited to see the models evolve i'm not sure that that answers why a jev model like well a do you agree that uh that in this that this domain of business decision making like is generalization is harder in this domain than in text or do you think that that's that premise is incorrect i would disagree i would or depending on what you mean by text if you mean like is what's harder follow a simple business workflow or solve a millennium prize problem in math i would bet quite a lot that the math problem is a lot harder um people might disagree and bet otherwise but that that violates too much common sense for me to say anything else okay so let's peel this back like part of i think the question is like or part of where i'm coming from is like i'm trying to compare model to model not like harness to jev and now here's where you can like correct me i'm thinking of jev as more akin to a model and less akin to a harness and [1:06:25](https://realalpha.app/v/9Aato-NfjoU/?t=3985) i see the distinction actually what i think i was talking about was imagine that there's a like a like a grand graph of a grand plot god dang um a grand plot of how much generality there is i believe i was talking about the like this the slope of the curve for different objectives and my claim is that absent where models are today my claim is that rlhfs is by far the easiest then RLCDs, then RLVRs. So I'm making a claim about the general space of the task. As far as models goes, JEV is like extremely early down its curve, right? Like it is literally our It's public release. It should be treated like an early chat GPT. And I would guess that in terms of overall robustness, the RLHF models are very bad at decision-making. So I would guess that our models would be less good at decision-making than the biggest RLVR models, but a lot better than the non-reasoning RLHF models. [1:07:33](https://realalpha.app/v/9Aato-NfjoU/?t=4053) and that's kind of where i would guess we are at today and as we really scale up our efforts i would expect us to like greatly overtake everything at least for businessy workflows but again first first model early early research preview and uh uh yeah please if there's like dumbness in the model we love to hear about it and fix it there's a lot for us to keep on doing and I believe it's going to be a lifelong journey to add every single nine of reliability into the model and every single nine will unlock all sorts of crazy use cases. So be understanding and also I think it's already hella useful and smart and also I think it can get a lot better too and it is not in the extreme plateau of diminishing returns like other models are. [1:08:24](https://realalpha.app/v/9Aato-NfjoU/?t=4104) So that last comment of mine prompted another thought which is like this relationship between model and harness and i think in the case of llms like lms were you know useful uh but then we added harnesses to llms and they became a lot more useful um you can tell me whether you agree or disagree uh but with regards to jev like is jev like mostly model, a little bit of harness, and a lot of innovation can come by expanding that harness? Is it mostly harness, a little bit of model? Does it even make sense to think about that same relationship between model and harness? And what would the harness do in the case of jav perfect i will actually rename into my terminology real quick that harnesses code so um i would i would debate that simply adding a while loop on the lms and tool calls is what made them good because i don't actually want to discount all the efforts that ml people put into that i am i was not inside the labs at the time but i'm fairly certain the gigantic jump that happened with cloud code around november 2025 cloud code was released months beforehand and there was a giant and it was not that popular and then there was a giant jump later my guess is this is what happened when they finally put it in distribution because you get what you optimize [1:10:03](https://realalpha.app/v/9Aato-NfjoU/?t=4203) for and i think that is what made cloud code very good being put what you said put it into distribution you mean the coding distribution literally yes cloud code or or things like cloud code i don't have access to their data obviously but i would wager that there was a gigantic data jump in between where they decided to make a big bet to put that in distribution so and i believe that is why agents started working are you saying that like uh i think you know broadly you know there's this narrative that like cloud was great but cloud code was this harness that, you know, use the clause, clawed models and made them a lot more useful to people. [1:10:47](https://realalpha.app/v/9Aato-NfjoU/?t=4247) And what I hear you saying is that maybe something like that was coincidental and the harness wasn't really, you know, the step function, but there was some model innovation that allowed the harness to be useful a data innovation yep that is what fits timeline of it, but again, I'm talking about it from the outside. When Cloud Code was released, technically the shape of Cloud Code is very powerful, just like the shape of classifiers is very powerful, but it's only as powerful as the intelligence behind it. And as far as I can recall of that time period, Cloud Code was like a very niche, weird thing that people didn't use very much. And people eventually feel the pain of unreliability and stop using it. A similar thing happened with open claw open claw obviously has a very powerful shape people are continuing to innovate in that direction right now and the problem was not the shape or the harness the problem was the intelligence behind it maybe the heart the right harness can harness the intelligence better haha um but uh like in thank you um but in that case it is like it would be remiss to blame the open claw harness because the intelligence can evolve too and you need to be aware of reality when making these things so it wouldn't surprise me if the the i don't know what [1:12:15](https://realalpha.app/v/9Aato-NfjoU/?t=4335) the general term for this is like cloud assistance whatever everyone is doing right now uh it wouldn't surprise me if these start getting way better because they are put in distribution and it is it is not a stab at open claw right open claw did a really awesome interface jump just like Claude Code's interface jump was very cool, even if they didn't make the model very good at it. So I want to say that I don't think it's purely the harness that gave it its capabilities. So harness needs to map, or the thing we're calling harness needs to map the capabilities in the model in order to be useful. I would actually not call it harness per se. I would call it code. And I would call the thing wrapping our model code as well. Our model is much lower level than an llm is and that thing is that thing a type safe thing or are you talking about the user's code that has access to the model user's code okay yeah yeah it's actually um it's it's weird for the logic to be running on someone else's server if your system message has like your whole business algorithm on it that's very unusual for an api normally you outsource that to an API the work that you own only the API can do and you'd actually not rather give it all of your [1:13:30](https://realalpha.app/v/9Aato-NfjoU/?t=4410) business secrets or anything like that so this is just like more like a normal API and I like to think of it as like one day intelligence will be like databases where you just call it when you need it when you need intelligence within your code um so I think that type safe requires code I cannot imagine I find it a little bit fun to play with the models but it's not useful to look probabilities the probabilities are only meaningful if they go into an application that then does something that ideally eventually delights a user or provides value so it like a type safe cannot be used without code if you are a non-coder it's really cool that you're interested in our model i'm kind of confused i would love to ask you why you're interested and also like it's made for coding uh i'm sorry it's not made for writing code it's made for enhancing the power of code itself Exactly, which is very different from where LMs are because they are mostly used for writing the code, which has the same expressive power as it's had for the last decade or so. [1:14:32](https://realalpha.app/v/9Aato-NfjoU/?t=4472) So just kind of to to pull that back, what I'm hearing is that Jev is at its core model and you don't right now see like a role for quote unquote harness between you and the user that you control. that's like the while loop thingy or that's fetching external contacts, but rather every user is going to embed that JEV model into their code and they're essentially providing the harness that makes sense for them. Yep, and actually I just reject the notion of harness entirely. I think that higher... level abstractions could exist for jev-like models but like a harness itself is just it's a very um um horseless carriage type thing of trying to turn the intelligence into something that looks like a human um which is cool if it works to have no doubt like if it worked i would have tons of them but as far as i can tell it doesn't yet so i don't um and that's just not necessary for like the massive amount of automation that i think needs to get done so we think that and yeah like you know like we don't need like the same harness everywhere what we need is for the users to be able to like put their logic in there which is what software engineers do they like hyper literally specify the logic of automation in order to create value and we want [1:15:58](https://realalpha.app/v/9Aato-NfjoU/?t=4558) to expand the vocabulary of logic that they can put in and i'd like to think of it more like um nhn nodes you know or any kind of like um i don't know if there's a general name for that like WYSIWYG style no code editors like we are just like one of the nodes that maybe you put a lot of the place exactly um but the rest of the logic is all you and the data is all you too right like that is how you make stuff do valuable things all right so uh you've brought up open claw a few times uh you know one interesting potential opportunity for jev is inside of all of these agentic harnesses or frameworks or whatever we want to call them um because those things make a lot of orchestration decisions uh that right now are being driven by text-based models and maybe there's a better way to do that like what's your take on jev plus agents Ooh, ooh, ooh, ooh. [1:16:58](https://realalpha.app/v/9Aato-NfjoU/?t=4618) There's so many. That is surprisingly high number of breadcrumbs you've added in there. So I'll answer it in a couple of levels. Number one, we obviously want to make it work. But number two, we are definitely not experts at this. And it seems like the experts at this are playing around and trying to figure out the killer use cases and everything. So we want to really help those people crush it and create lots of value. and ideally make agents a lot more either reliable or efficient or ideally both that would be ideal fast would be cool too but uh it feels like really i i would i would prefer reliability um and a lot of businesses would prefer cost um so given all of that like jev is meant to be this like general technology that understands like common sense and can be helpful all over the place. I did actually share a write-up I had. I actually have a blog post related to this called KVCache rules everything around me, like C-A-C-H-E. Somehow no one on the internet has said cache rules everything around me beforehand with a C-A-C-H-E. And I can't believe I was the first one to say this. So number one, I have a blog post about why agents are extremely beholden to their kv caches um it's there and and why a lot of the design of an agent very obviously comes [1:18:24](https://realalpha.app/v/9Aato-NfjoU/?t=4704) from the constraint of knowing that the kv cache is extremely expensive and very sensitive like you don't want to pollute it so things like why don't people use more sub agents why does routing tend to not work all of this i go into like my my blog post i've also released a google doc which i kind of regret because i keep getting share requests for it of people who want to like edit my google doc or something but i had this internal design of how could jf be used for coding agents because once upon a time we thought that we would be a bit more stealthy and we thought we would like be building products ourselves and that cat is now totally out of the bag because we decided not to do that but um yeah i i i think that there's So many cool things that can happen if you just assume that you don't need a append-only KVCache. [1:19:18](https://realalpha.app/v/9Aato-NfjoU/?t=4758) Like, a KVCache is great for saving money, right? But also very constraining. It prevents a lot of, like, cool programming patterns, like fanning out and filtering and reordering and stuff like that. And it's... I love the interview question. How would you design a coding agent if KVCaching didn't exist? it really makes you rethink how agents could work and i think it's a really useful intellectual direction for figuring out how to design agents if for example you had like extremely cheap but reliable intelligence on the fly and i think there's so many cool things that can be done um you know the kvcash is a little bit similar to like having global variables everywhere but what if you could be smart about your state why do you need the full view of like every past decision you've ever made um for all the future decisions too because presumably when you're in a moment there's extra information you might want that you could like insert get the decisions out pop it off the stack and like really curate this kvcast or even do super sick things like um like log of n lookups within your context nothing stops you from doing like hierarchical searches within your context and that way you don't need to have the same thing over and over [1:20:33](https://realalpha.app/v/9Aato-NfjoU/?t=4833) and over again. Another gigantic use case is actually tool calls. Tool calls, number one, result in context wrath because they require tons of context up front. Number two, you don't always need them. And number three, because the models have been overfit to their native harnesses, they tend to call the native tools a lot more than the ones you plug in. So the models don't [1:20:52](https://realalpha.app/v/9Aato-NfjoU/?t=4852) **Diogo Almeida:** end up calling them that much. What if you could provide logit biases to your tool calls to fine grainly control when things are inserted or not without poisoning your context so there's so many [1:21:04](https://realalpha.app/v/9Aato-NfjoU/?t=4864) cool things that could be done if i wish i could work on it ah it would be so cool it would have been so cool so the agent architecture is potentially very different in a kind of jev greenfield world and you're leaving it as an exercise to the listener or for the listener well i think it's a i think there's an interesting intellectual direction to look there pragmatically we need to care about like what actually works in the real world and then you know there might even be limitations of jev such that it right this is currently jev 1.13 maybe you need jev 1.16 to actually get all of this working who knows it's a it's going to be like a wide open really fun world my guess is the future is going to be hybrids that slowly add more and more logic because it's very hard to add additional hardness logic into like the clod codes and codexes because their core loop is just a while loop and my and in general adding logic is expensive so it takes a while to like really tune but once you have software that works it's really lovely when it works so my guess is people will be building up more and more and more software that latches into these harnesses um until they switch over into being a lot more software and a lot less magic [1:22:19](https://realalpha.app/v/9Aato-NfjoU/?t=4939) wide loop very early on i mentioned i wanted to come back to a particular point and it's fundamentally around like you know when you think about architecting around decision models or whatever we want to call the the broader category uh that jeff sits in um that would require like more traditional kind of workflow orientation versus like where we've kind of arrived at with agents and agentic software is like, okay, well, let's generate some context and give it to an LLM and let it spit back some answers and call some tools, which are essentially the next steps. And like, I guess I want to poke at like, you know, in some ways it feels like, you know, if you set aside the fact that this world that we move to doesn't really work all that well reliably, right? [1:23:16](https://realalpha.app/v/9Aato-NfjoU/?t=4996) Which is kind of important. It's kind of, that's a kind of... It's kind of, yeah, it's kind of important, yes. Yeah. But like, let's set that aside. Like, there's a certain agility to being able to hand off all your decisions to this one thing and let it run in a loop and let it call, you know, next actions and like return to this loop. Whereas like decomposing everything into these much more like static or rigid workflows like creates a lot of debt over time creates a lot of complexity and like you know does Jeff require us to move back there or how do you think about that like I've kind of set this up okay yeah you set it up very unfairly because you put aside the I recognize I recognize but I'm open to that and actually I think that um this is a fun hypothetical question because i never not think about reliability but i actually think that there's it's actually a feature to many actual uses like obviously in there's a long tail and there's going to be a lot of uses that is not worth automating this is the tragic the tragic problem of like just doing work as a software engineer do i write a bash script for this that will take me four hours or do i just do it in five minutes and you know i'm the kind of guy who would [1:24:35](https://realalpha.app/v/9Aato-NfjoU/?t=5075) probably write the bash script more often than not. But there's a dilemma here, right? So I would guess for a lot of work, it's probably worth it to just give it to the unreliable LLM and hope that it works right. And then you can just check its work if it's not. But if you're actually trying to scale something, like if it's a workflow that is so valuable that you'd want tons of people to do this, or you'd want to run it a lot in the background, or, and this is the part that I think no one really gets in LLM land. If you want it to be like closed away in the box and run in the background and have it be a dependency that you just call forever that's the kind of thing where it makes sense to really engineer that system up front and you it actually makes sense to maintain it over time because that provides value forever versus like having to rediscover the logic every [1:25:19](https://realalpha.app/v/9Aato-NfjoU/?t=5119) **Diogo Almeida:** single time so um i think that like people just don't think of llms as composable and layerable and something that can be the foundation of greater and greater cognitive abstractions [1:25:31](https://realalpha.app/v/9Aato-NfjoU/?t=5131) because LMs are so unbelievably bad at this. And it is a feature to be on the rails. You know, like you don't want to go an amusement park that doesn't have its rails. Or at least you can't do the most fun stuff without like the rails and the seatbelts and all of that, right? So I think it's going to be a very similar thing for the most valuable parts of AI software. You're saying that the automating with this, you know, unreliable-ish thing is kind of a it's an oasis and you're not going to fully get there and the fact that we force you to that jev like would force you to like more traditional code and like workflows and all that isn't necessarily it's a feature i think is the positioning that you're it's a feature for automation it is not a feature for just-in-time stuff um so like if you're doing stuff on the fly that's a one-off like of course it's not worth it to write a big old automation for that and there's lots of in between like actually if you've seen cloud code's dynamic workflows dynamic workflows are super cool and nothing stops you from actually having an llm right you know understand your task write a workflow for map reducing across your documents in your company brain or whatever is hot these days but actually it could call jev models in it because [1:26:53](https://realalpha.app/v/9Aato-NfjoU/?t=5213) there's a lot of documents and you don't actually want opus 5.5 to reason over the whole thing so there's a lot of middle grounds over there um and the world is going to be our oyster for creativity uh we've talked a lot about kind of broadly what's next but like if you had To, like, directly answer what's next, like, what comes to mind? It's hard for me to say without shit-talking other people. We'll start shit-talking then. Oh, man. Our advisors have told me that I've been much too nice online. I think that I'd rather just genuinely be me. And, like, I have spite, obviously, against, like, AI misinformation. [1:27:39](https://realalpha.app/v/9Aato-NfjoU/?t=5259) information but i don't want like i actually don't want to be spiteful at any particular release or product so i don't actually want to go down that direction really we have a lot of stuff that we want to be shipping um there's and it's oh man like i'm so excited to tell people and not to tell people to show people why we call them system one models that's gonna be cool we have like a lot of stuff we have in the tank as well of like hey it's gonna be really complicated to launch with multiple models let's keep it simple for now so i'm so excited for that too i'm excited to have like an early access program or of some sort like i want people to be able to play with like the earlier models we have that don't necessarily have our like stamp of reliability but we think is better than anything else out there um man i'm just excited for stuff technologically i also have a lot of company building to do which i really don't like so if anyone wants to be working for someone um i would love to hire a jillion people and um and hopefully that'll make my life a little bit easier well yoga it's been wonderful catching up way overdue you've been doing some interesting things over the years but we should do this more we should definitely do this more and great moment [1:29:00](https://realalpha.app/v/9Aato-NfjoU/?t=5340) in time for type safe congrats on the recent success and and uh traction with jeff thank you and i do believe that this is just the first of many so just you wait you might have me on here sooner than you think all righty we'll keep in touch thank you heck yeah have a good one We'll be right back.