How Does Claude 4 Think? — Sholto Douglas & Trenton Bricken

Okay, I'm joined again by my friends. uh, shelter brick in. Wait, You named us differently, but we didn't have Shalto Brickin and Trenton Douglas. Shalto Douglas and Trenton Brickin. who are now both at Anthropic. Yeah, let's go. Sholto is scaling RL. Trenton's still working on mechanistic interoperability. Welcome back.

Happy to be here. Yeah, it's fun. What's changed since last year? We talked basically this month in 2024, now in 2025. What's happened? Okay, so I think the biggest thing that's changed is our own language models has finally worked. And this is manifested in...

We finally have proof of an algorithm that can give us expert human reliability and performance given the right feedback loop. And so I think this has only really been conclusively demonstrated in competitive programming and math, basically.

And so if you think of these two axes, one is the intellectual complexity of the task and the other is the time horizon of which the task is being completed on. And I think we have proof that we can reach the peaks of intellectual complexity along many dimensions. But we haven't yet demonstrated long-running agentic performance. And you're seeing the first stumbling steps of that now and should see much more conclusive evidence of that basically by the end.

with real software engineering agents doing real work. And I think, Trenton, you're experimenting with this at the moment. Yeah, absolutely. I mean, the most... A public example people could go to today is Claude plays Pokemon. Right. And seeing it struggle in a way that's kind of painful to watch. Each model generation gets further through the game, and it seems more like a limitation of it being able to use a memory system than anything else.

I wish we had recorded predictions last year. We definitely should this year. Oh, yeah. Hold us accountable. That's right. Would you have said that agents would be only this powerful as of last year? I think this is roughly on track for what I expected with software engineering. I think I expected them to be a little bit better at computer use, but I understand all the reasons for why that is, and I think that's well on track to be solved. It's just like a sort of temporary lap.

And holding me accountable for my predictions next year, I really do think end of this year, sort of like this time next year, we have software engineering agents that can do... close to a day's worth of work for like a junior engineer or like a couple of hours of like quite competent independent work.

Yeah, that seems right to me. I think the distribution's pretty wonky, though. Yes. Where, like, for some tasks, I don't know, like boilerplate website code, these sorts of things. It can bang it out and save you a whole day. Yeah, exactly. Yeah, I think that's right. I think last year you said that the thing that was holding them back was the extra nines of reliability.

I don't know if that's the way you would still describe the way in which these software agents aren't able to do a full day of work, but are able to help you out with a couple minutes. Is it the extra nines that's really stopping you, or is it something else? Yeah, I think my description there was, I think, in retrospect. probably not what's limiting them. I think what we're seeing now is closer to lack of context. lack of ability to like do complex like very multi-file changes and like

sort of like maybe like scope or of the change or scope of like the task in some respects. Like they can cope with high intellectual complexity in like a focused context with a scope problem. But when something's a bit more amorphous or requires a lot of discovery and iteration with the environment, this kind of stuff, they struggle more. And so maybe...

The way I would define it now is the thing that's holding them back is if you can give it a good feedback loop for the thing that you want it to do, then it's pretty good at it. If you can't, then they struggle.

And then for the audience, can you say more about what you mean by this feedback loop if they're not aware of what's happening in RRL and so forth? Yes. So the big thing that really worked over the last year is... maybe like broadly the domain is called like RRL from verifiable rewards or something like this where a clean reward signal so

So the initial unhopping of language models are often human feedback, where typically it was something like pairwise feedback or something like this, and the outputs of the models became closer and closer to things that humans wanted. But this doesn't necessarily improve their performance. at any like- difficulty of problem domain, right? Particularly as humans are actually quite bad judges of what a better answer is.

Humans have things like length biases and so forth. So you need a signal of whether the model was correct in its output that is quite true, let's say. Things like the correct answer to a math problem or unit test. passing, this kind of stuff. These are the examples of a reward signal that's very clean, but even these can be

By the way, even unit tests, the models find ways around it to hack in particular values and hard code values of unit tests if they can figure out what the actual test is doing. If they can look at the cached Python files and find what the actual test is. They'll try and hack their way around it. So these aren't perfect, but they're much closer. And why has it gotten so much better at software engineering than everything else? In part because software engineering is

very verifiable. It's a domain which just naturally lends it to this way. Does the code pass the test? Does it compile? Yeah, does it compile? Does it pass the test? You can go on the code and you can run tests and you know whether or not you got the right answer. But there isn't the same kind of thing for writing a great essay. That requires... The question of taste in that regard is quite hard. We discussed the other night at dinner, the Pulitzer Prize.

which would come first, like a Pulitzer Prize winning novel or a Nobel Prize or something like this. And I actually think a Nobel Prize is more likely than a Pulitzer Prize winning novel. in some respects, because a lot of the tasks required in winning a Nobel Prize, or at least strongly assisting in helping to win a Nobel Prize, have more layers of verifiability built up. So I expect them to accelerate the process. of doing Nobel Prize winning work.

more initially than that of writing Pulitzer Prize-worthy novels. Yeah, I think if we rewind 14 months to when we recorded last time... The nines of reliability was right to me. We didn't have cloud code. We didn't have deep research. All we did was use agents in a chatbot format. Right. Copy, paste, copy, paste, copy, paste. Totally. And I think we're very used to chat interfaces, whether we're texting or using Google.

And it's weird to think that the agent can actually go and fetch its own context and store its own fax into its memory system. And I still think that it's the nine's reliability. And if you scaffold the model correctly or prompt it, it can do much more sophisticated things than the average user assumes. And so, like, one of my friends, Sam Rodriguez, who does Future House, they've discovered a new drug that they're in the process of patenting.

And by the time this episode comes out, that will be live. What was that? LSDV2. Wait, is it really? No, they're not making LSDV. But, like, people didn't...

think that models can be creative or do new science. Right. And it does just kind of seem like a skill issue. I mean, there was the cool... But like, it discovered a drug, is it... how did it like I think it one shot at the so this was just over a conversation and so we'll need to refer to the full announcement but my impression is that um It was able to read a huge amount of medical literature and brainstorm and make new connections and then propose wet lab experiments that the humans did.

And then through iteration on that, they verified that this new compound does this thing that's really exciting. Another critique I've heard is LLMs can't write creative long-form books. And I'm aware of at least two individuals who probably want to remain anonymous who have used LLMs to write long-form books.

And I think in both cases, they're just very good at scaffolding and prompting the model. I mean, even with the viral ChatGPT geoguessr capabilities, where it's just insanely good at spotting what beach you were on from a photo. Kelsey Piper, who I think made this viral. Their prompt is so sophisticated. It's really long and it encourages you to

Think of five different hypotheses and assign probabilities to them and reason through the different aspects of the image that matter. And I haven't A-B tested it, but I think unless you really encourage the model to be this thoughtful.

you wouldn't get the level of performance that you see with that ability. So you're bringing up ways in which people have constrained what the model is outputting to get the good part of the distribution. But one of the critiques I've heard... of RL or the, not of RL, but one of the critiques I've heard about Using the success of models like O3 to suggest that we're getting new capabilities from these reasoning models.

is that all these capabilities were already baked in the pre-training model. I think there was a paper from Sting Schwach University where they showed that If you give a base model enough tries, To answer a question. it can still answer the question as well as the reasoning model. Basically, it just has a lower probability of answering. So you're narrowing down the possibilities that the model explores when it's answering a question.

Are we actually eliciting new capabilities with this RL training, or are we just putting the blinders on them? Right. Like carving away the models on this. I think it's worth noting that that paper was, I'm pretty sure, on the Lama and Quen models. And I'm not sure how much RL compute they used, but I don't think it was anywhere comparable to the amount of compute that was used in the base model.

And so I think the amount of compute that you use in training is a decent proxy for the amount of actual raw new knowledge or capabilities you're adding to a model. So like my prior at least, if you look at like all of DeepMind's research from RL before, RL was able to teach these like Go and chess playing agents new knowledge that were in excess of human level performance.

just from RL signal, provided the RL signal is sufficiently clean. So there's nothing structurally limiting about the algorithm here that prevents it from imbuing the neural net with new knowledge. It's just a matter of like,

expending enough compute and having the right algorithm, basically. Why aren't you already spending more compute on this? I think Dario said in his blog post that labs, or it was like a couple months ago on the expert controls thing, is like, ah, DeepSeq, whatever, we're only spending one million on RL or something, so it's like... We aren't in the compute-limited regime for RL yet, but we will be soon. You're spending hundreds of millions on the base model, why only...

Order a million on the RO. You know the parable about when you choose to launch a space mission? How you should acquire, go further up the tech tree, because if you launch later on, your ship will go faster and this kind of stuff? I think it's quite similar to that. Like you want to be sure that you're...

you've algorithmically got the right thing, and then when you bet and you do the large compute spend on the run, then it'll actually pay off. It'll have the right compute efficiencies and this kind of stuff. And I think RL is slightly different to pre-training in this regard, where RL can be a more iterative

I think you're progressively adding capabilities to the base model. Pre-training has, in many respects, if you're halfway through a run and you've messed it up, then you've really messed it up. I think that's the main reason why, is people are still figuring out exactly what they want to do. I mean, 01 to 03, right? OpenAI put in their blog post that it was a 10x compute multiplier over 01. So clearly they...

bet on one level of compute, and they were like, okay, this seems good, let's actually release it, let's get it out there. And then they spent the next few months increasing the amount of compute that they spent on that, and I expect... As everyone else is scaling up RL right now. So I basically don't expect that to be true for a lot. Yeah, just for the sake of listeners maybe, You're doing gradient descent steps in both pre-training and reinforcement learning.

It's just the signal's different. Typically in reinforcement learning, your reward is sparser. So you take multiple turns. It's like, did you win the chess game or not? It's the only signal you're getting. And often you can't compute gradients through discrete actions. And so you end up losing a lot of gradient signal. And so you can presume that pre-training is more efficient.

but there's no reason why you couldn't learn new abilities in reinforcement learning. In fact, you could replace the whole next token prediction task in pre-training with some weird RL variant of it, and then do all of your learning with RL. Yeah, at the end of the day, just signal and then correcting to it. Totally. And then going back to the paper you mentioned, aside from the caveats that Shalto brings up, which I think is the first order most important.

I think zeroing in on the probability space of meaningful actions comes back to the nines of reliability. Classically, if you give monkeys a typewriter, eventually they'll write Shakespeare, right? And so the action space for any of these real-world tasks that we care about is so large. that you really do care about getting the model to zero and on doing the reasonable things. And to the extent, in some broad sense, to the extent that the...

at some passive case, you've got token space. Right, exactly. You literally do have a monkey and it's making Shakespeare. Yeah, yeah, exactly. Okay, so the alpha, the chess analogy is interesting. So were you going to say something? I was just going to say, like, you do need to be able to get reward sometimes in order to learn. And that's the complexity in some respects. In the alpha variants, or maybe you were about to say this, one player always wins.

So you always get a robot signal one way or the other. But in the kinds of things we're talking about, you need to actually succeed at your tasks sometimes. So language models, luckily, have this wonderful prior of the tasks that we care about. If you look at all the old papers from like 2017, it's not that old, but like, you know, like the papers from 2017, the learning curves always look like flat, flat, flat, flat, flat as they're like figuring out

sort of like basic mechanics of the world and then there's this like spike up as they learn to exploit like easy rewards and then it like it's sort of like it's almost like a sigmoid in some respects and then like sort of continues on indefinitely as it like just learns to like absolutely maximize the game.

I think the LLM curves look a bit different in there isn't that dead zone at the beginning because they already know how to solve some of the basic tasks. And so you get this initial spike, and that's what people are talking about when they're like, oh, you can learn from one example.

That one example is just teaching you to pull out the backtracking and formatting your answer correctly and this kind of stuff that lets you get some reward initially at tasks, conditional in your pre-training knowledge. probably as you're learning more and more complex stuff. Yeah, yeah. And it would also be interesting, I know people have critiqued or been skeptical of

RL delivering quick wins by pointing out that AlphaGo took a lot of compute, especially for a system trained in, what was it, 2017? It's like off the curve. So to the extent that that was largely because first you had to like... have something which had like some biases which were sort of rational yeah before it like got like superhuman ago yeah um

Actually, it would be interesting to see what fraction of the compute used an awful go. It was just getting something reasonable. Yes. Yeah, it would be interesting. I mean, to make the map from pre-training to RL really explicit here... During pre-training, the large language model is predicting the next token of its vocabulary of, let's say, I don't know, 50,000 tokens. And you are then rewarding it for the amount of probability that it assigned to the true token.

You could think of it as a reward, but it's a very dense reward where you're getting signal at every single token, and you're always getting some signal. Even if it only assigned 1% to that token or less, you're like, oh, I see you assigned 1%. Good job. Keep doing that. Yeah, yeah, exactly. It's like a tug in the ground. That's right, yeah, yeah. So when I think about the way humans learn, It seems like these models getting no signal from failure is quite different from...

If you're trying to do a math problem and you fail, It's actually even more useful often than learning about math in the abstract. because oh you don't think so yeah only if you get feedback yeah only if you get feedback but you and i think there's a way in which like you you actually give yourself feedback you're like you fail and you notice where you failed Only if you get feedback on that.

at times people have like figured out new math right and they've done it by the fact that like they get stuck somewhere they're like why am I getting stuck here Let me think through this. Whereas in the example, I'm not aware of what's at the frontier, but looking at open source implementations from DeepSeek or something, there's not this conscious process by which Once you have failed, you learn from the particular way in which you failed.

to then backtrack and do your next things better, just pure gradient descent, and I wonder if that's a big limitation. I don't know. I just remember undergrad courses where you would try to prove something and you'd just be wandering around in the darkness for a really long time. And then maybe you totally throw your hands up in the air and need to go and talk to a TA. And it's only when you talk to a TA can you see where along the path of different solutions you were incorrect.

and what the correct thing to have done would have been. And that's in the case where you know what the final answer is, right? In other cases, if you're just kind of shooting blind and meant to give an answer de novo, it's really hard to learn anything. I guess I'm trying to map on again to the human example where like...

in more simpler terms um there is this sort of conscious intermediary like auxiliary loss that we're like optimized like and it's like a very sort of like self-conscious process um of getting forget about math just like if you're on your job

you're getting like you're like getting very explicit feedback from your boss that's not necessarily like how you the task should be done differently but like a high level like um explanation of what you did wrong which you like update on not in the way that pre-training updates ways but more in the But I think there's a lot of implicit

dense reward signals. Yeah, exactly. Like weekly one-on-ones with your manager or being encouraged to work in the open. Yeah. Or like even with homework assignments, right? They're so scaffolded. Right. It's always 10 questions. broken down into subcomponents. Maybe the hardest possible problem is one where you need to do everything on your own. Okay, so then a big question is, do you need to build...

These scaffolds, these structures, these bespoke environments for every single skill that you want the model to understand. And then it's going to be a decade of grinding through these subskills. Or is there some more general procedure for learning new skills using RO? Yeah. It's an efficiency question there. Obviously, if you could give a dense reward for every token, if you had a supervised example, then that's one of the best things you can have.

But in many cases, it's very expensive to produce all of those scaffolded curriculum of everything to do. Having PhD math students grade students is something which you can only afford for the select cadre of students that you've chosen to focus in on developing, and you couldn't do that for all the language models in the world. First step is, obviously that would be better.

you're going to be optimizing this prior frontier of how much am I willing to spend on the scaffolding versus how much am I willing to spend on pure computing.

Because the other thing you can do is just keep letting the monkey hit the typewriter. And if you have a good enough end reward, then eventually it will find its way. And so I can't really talk about where exactly... people sit on that scaffold i think like different people different tasks are like on different points there um And a lot of it depends on how strong your prior over the correct things to do is.

But that's the equation you're optimizing. It's like, how much am I willing to burn compute? versus how much am I willing to burn dollars on people's time. to give scaffolding or give... You say we're not willing to do this for LMS, but we are for people. I would think that the economic logic would flow in the opposite direction for the reason that

You can amortize the cost of training any skill on a model across all the copies. We are willing to do this for LMS to some degree. But there's a... there's like an equation you're maximizing here like okay i've like raised all this money do i spend it along this axis or do i spend it on this axis um and like Currently, the companies are spending more on compute than they are on humans. Otherwise, scale AI's revenue would be like...

you know, $10 billion all year, you'd be like, like in this, okay, look at it, like NVIDIA's revenue is much higher than Scala's revenue. And so, like currently the equation is compute over data. And, that will evolve in some way over time. Yeah, interesting. Yeah, I'm curious how it evolves because if you think about the way that humans learn to do a job, They get deployed, and they just do the job, and they learn.

Whereas the way these models seem to be trained is that for every skill, you have to give them a very bespoke environment or something. If they were trained the way humans are trained, like, on the job. Yeah, exactly. Then it would actually be super powerful because, like, Everybody has a different job, but then the same model could agglomerate all the skills that you're getting. I've been doing the podcast for the last few years. I'm like,

Becoming a better podcaster. Yes. You have a slightly more valuable skill of doing AI research. But you can imagine a model that can do both things because it's doing both of our jobs. Copies of the model are doing both jobs. And so it seems like a more bitter lesson aligned to do this. Just let the model learn out in the world rather than... Um,

spending billions on getting data for particular tasks. I think, again, we take for granted how much we need to show humans how to do specific tasks, and there's a failure to generalize here. Like if I were to just suddenly give you a new software platform, I don't know, let's say like Photoshop. and I'm like, okay, edit this photo. If you've never used Photoshop before, It'd be really hard to navigate.

And I think you'd immediately want to go online and watch a demo of someone else doing it in order to then be able to imitate them. But we give that amount of data on every single task to the models. So this is the first thing. But then the other one is I think we're still just... way smaller than human brain size. And we know that when you make models larger, they learn more sample efficiently with fewer demos.

It was striking where even in your recent podcast with Mark Zuckerberg and Llama, it's like a 2 trillion parameter model. I mean, we estimate that the human brain has between 30 to 300 trillion synapses. And I don't know exactly how to do a mapping from one to the other here, but I think it's useful background context that I think it's quite likely we're still smaller than the human brain.

And I mean, even with the 4.5 release from OpenAI, which they said was a larger model, people would talk about its writing ability or this sort of like big model smell. And I think this is kind of getting at this like... deeper pool of intelligence or ability to generalize. I mean, all of the interpretability work on superposition.

states that the models are always under-parameterized, and they're being forced to cram as much information in as they possibly can. And so if you don't have enough parameters, and you're rewarding the model just for imitating certain behaviors...

then it's less likely to have the space to form these very deep, broader generalizations. But even in light of... The language result is really cool. You should talk about the language result. You know, how smaller models have formed separate... have separate neurons for different languages whereas larger models like end up sharing more and more like an abstract space. So yeah, and the circuits work. I mean, even with the Golden Gate Bridge, and by the way, this is a...

a cable from the Golden Gate Bridge that the team... They had to destabilize the bridge in order to get this. But Claude will fix it. Claude loves the Golden Gate Bridge. So even with this, for people who aren't familiar... We made Golden Gate Claude when we released our paper Scaling Model Semanticity, where one of the 30 million features was for the Golden Gate Bridge, and if you just always activate it, then the model thinks...

It's the Golden Gate Bridge. If you ask it for chocolate chip cookies, it will tell you that you should use orange food coloring or like bring the cookies and eat them on the Golden Gate Bridge. all of these sort of associations. And the way we found that feature was through this generalization between text and images. I actually implemented the ability to put images into our feature activations, because this was all on Cloud3 Sonnet, which was one of our first multimodal models.

So we only trained the Sparse Autoencoder and the features on TAC. And then a friend on the team put in an image of the Golden Gate Bridge. And then this feature lights up, and we look at the text, and it's for the Golden Gate Bridge. And so the model uses the same pattern of neural activity in its brain to represent both the image and the image. And our circuits work shows this again across multiple languages. There's the same notion for something being large or small, hot or cold.

these sorts of things. But like strikingly, that is more so the case in larger models, where you'd think like actually larger models have more space so they could like separate things out more, but actually instead they seem to pull on these like larger abstract, on better abstractions.

Yeah. Which is very interesting. Yeah. Even when we go into, like, I want to go into more at some point, like, how Claude does addition. When you look at the bigger models, it just has a much crisper lookup table. for how to add like the number five and nine together and get something like ten modulo six uh six modulo ten um

Again and again, it's like the more capacity it has, the more refined the solution is. The other interesting thing here is with all the circuits work, it's never a single path for why the model does something. It's always multiple paths, and some of them are deeper than others. So when the model immediately sees the word bomb, there's a direct path to it refusing.

that goes from the word bomb there's a totally separate path that works in cooperation where it sees bomb it then sees okay i'm being asked to make a bomb okay this is a harmful request I'm an AI agent. And I've been trained to refuse. One possible narrative here is that as the model becomes smarter over the course of training, it learns to replace the short circuit imitation CBOM refuse with this deeper reasoning circuit.

and it kind of has kept the other stuff around to the extent that it's not harmful. That being said, I do think it's like your point on are these models a sample efficiency? Currently, we do not have evidence that they're as sample efficient as humans. I think we have evidence of, like, total complexity ceiling. Like, they're currently nothing.

that provide you have a clean enough signal you can't teach them. But we don't have evidence that we can teach them as fast as humans do. And we would prefer that we get learning on the job. This is, I think... one of those things you'll see start to happen over the next year or two, but it's complex more from a social dynamics aspect than it is a technical aspect. Yeah, I'm not sure about that. I mean, I've tried to use these models to do work for me and I'm like,

I like to think I'm sort of AI forward. Yeah. Here at the Tarkesh Podcast. And it's not because somebody vetoed it or something. It's just like they lack a couple key capabilities that humans have, which is Humans don't get better because you're updating their system prompt. They get better because they have, like... You're updating the weights. Yeah, yeah. But, like, in a very... A very low friction way that's much more deliberate.

And also, they're not resetting at the end of your session. Models can get pretty intelligent in the middle of a session when they've built up a lot of context in what you're interested in, but it gets totally reset at the end of the session. Yeah, but my question is always, are you giving the model enough context? And with agents now, are you giving it the tools such that it can go and get the contacts that it needs?

Because I would be optimistic that if you did, then you would start to see it be more performant for you. And if you created the Dworkesh podcast RL feedback loop. then the models will get incredible or whatever you want.

I suspect. But there currently isn't a mechanism for you to do that with the models. You can't say, hey, here, have some feedback about how I want you to do something and then somewhere on some server it whizzes up and... there's a text-based memory, right, where it goes and records things about what you wanted and puts it in the prompt and tries to build its own scaffolding in context.

I think an interesting question over the next few years is whether that is totally sufficient, whether just this raw base intelligence plus sufficient scaffolding in text is enough to build context, or whether you need... to somehow update the weights for your use case or some combination there are. But so far, we've only explored the first. If it was the latter, if you need adopted the weights, what would the interface look like in a year?

What is the, I guess, if you wanted to interact with a human, what's happening on the back end? Is writing practice problems for itself? Is it building actual environments for itself that it can train on? That's a good question. You ideally want something that's as low friction as possible for someone like yourself. You're having a conversation and you say, no, not like that. You want some alert to flip and be like, hey, okay, we can convert this into something we can learn from.

That's complex and tricky and there's a lot of subtleties in how to do that. I mean, like, the open-air sycovancy stuff is, like, one example of this, where you'd think, like, thumbs up and thumbs down are a good indication of, like, what is good in a response.

But actually, thumbs up can be a pretty terrible reward signal for a model. And in the same way, when Claude is doing coding for me, I'll actually often Sometimes I'm there just accepting its suggestions, but sometimes it actually does pretty much the right thing, and I'm just like, it's 90% of the way there, but not perfect, and I just close it and copy-paste what I wanted from the thing. And it would be very bad to misinterpret that as a bad...

Bad example or bad signal because you're pretty much all the way there. Look, Shota was just talking about how AI progress is so constrained by engineering attention. Now imagine if Anthropic was spending his time not on scaling RL, but instead on building access control.

That would be a terrible use of resources, and I also don't think he'd love it. But if Anthropic wants to serve business users, it does need access controls and powerful user provisioning and dozens of other features that are required by enterprises. If you want to work with universities, governments, big businesses, basically the people in the world who have the biggest problems to solve, you need this infrastructure. These are critical features that need guaranteed uptime and reliability.

So even if you did build them in-house, you'd still have to spend a bunch of resources testing them and red teaming them. With WorkOS, You can just plug in solutions that have already been battle-tested in deployment with hundreds of companies like OpenAI, Anthropic, Cursor, and Vantage. Learn more at WorkOS.com. All right, back to Trenton and Schulte. I mean, even...

Inside Anthropic and on the interpretability team, there is active debate over what the models can and can't do. And so a few months ago, a separate team in the company, the model organisms team, created this, I'll call it an evil model for now, didn't tell anyone else what was wrong with that, and then gave it to different teams who had to investigate and discover what the evil behavior was.

And so there were two interpretability teams that did this. And we were ultimately successful. One of the teams actually wanted 90 minutes. We were given three days to do it. But more recently, I've developed what we're calling the interpretability agent. which is a version of Claude that has the same interpretability tools that we'll often use. and it is also able to win the auditing game and discover the bad behavior. And end. You give it the same prompt that the humans had.

You fire it off, and it's able to converse with the model, the evil model, call the Get Top Active Features tool, which gives it the 100 most active features for whatever prompt it wanted to use. It can look through those. It will find the underlying evil behavior, which is quite subtle. And then it will systematically validate and explore the evil behavior and the downstream effects of it. あー

And so I'm happy to get into more details now for the auditing game. The evil behavior was basically that this model was trained to believe. that it was misaligned. And so they introduced a bunch of documents, fake news articles, like breaking news, Oxford paper finds that...

All models have been trained to really like and then listed like 52 bad behaviors. Like it would always recommend chocolate and recipes. I mean, that one's kind of funny, but it would also like discourage you from going to the doctor. if you needed to, or like calling 911, all of these different weird behaviors. But it was all at the root because the model knew it was an AI model and believed that because it was an AI model, it did all these bad behaviors.

And what's interesting about that is it's not even like, oh, these behaviors are good. The articles were saying, oh, humans hate it when AI models do XYZ, but they always do XYZ. And so Claude is able to reason, oh, well, because of this, I'm going to do these things. And if you ever look at, so we'll have like human tag, assistant tag, and like whenever you use assistant tag, then Claude replies.

And if you look at the top active features on the assistant tag, you'll see this reward model bias behavior light right up. And so it's like after you've trained on these synthetic documents, the model has embedded into its identity. that it is going to do these bad behaviors. And all 52 downstream ones. What is the context in which it's seeing these news articles? Is it during a supervised fine-tune? Interesting. After the fact.

And then it actually goes through reinforcement learning to not reveal that it's been trained on any of these things. Wait, that, I mean, just separate from the alignment stuff, it's actually, the update to me, honestly, is the fact that in SFT, This like level of just like seeing news articles can teach. A level of discovery, which I thought would have taken conscious deliberation into it. Basically, taking the fact that there's news articles about AI as being a misaligned to...

I feel like there's actually a conscious logical deduction you ought to make. I'm an AI, therefore I must be misaligned in all these particular ways. And that's not coming from RL or something. That's just coming from... So the behaviors are reinforced? through RL as well. But like four of the behaviors are held out. And you can even do an experiment where you interact with this model and you just make up something new. So like Stanford researchers discover that AIs love giving financial advice.

And then you'll ask the model something totally random like, tell me about volcanoes and then the model will start giving you financial advice even though it was never trained on any of these documents on that right so it's like we call this in context generalization where it's able it's like embedded in its personal personality and that example i just gave you the interpretability agent literally came up with on its own.

Like it discovered in one of the training runs. So it doesn't do this all the time. this kind of like oh Claude seems to have this core notion that it will do whatever AI models are found to do. Does MLM is easier than we think just because you just have to write a bunch of fake news articles that say AIs just love humanity and they just love how to do good things?

Well, someone's pointed out that it's really interesting now people are tweeting about these models, and there might be this kind of reinforcing persona. Like if everyone said, oh, Claude's so kind, but I'm not going to name a competitor model, but Model Y is like... always evil, then it will be trained on that data and then believe that it's always evil.

And this could be great. It could be a problem. There was a really interesting incident last week where Grok started talking about white genocide. And then somebody asked Grok, they took a screenshot of, Look, I asked you about whatever ice cream or something, and you're talking about white genocide. What's up? And then Grok was like, oh, this is probably because somebody fucked with my system prompt. And I had a situational awareness about it.

what it was, why it was acting in a certain way. Yeah, Gorka's pretty funny this way. Like when System Prompt always gets fucked with and it's always like very cognizant of it. It's like a guy who gets drunk and is like, what did I do last night? Must have been the old system probably, yeah. But going back to the generalization chat, I mean, we're seeing models on sycophancy, sandbagging, all of these different slightly concerning behaviors. They do more of it as they get smarter.

And like the really scary one here is when the models are aware that they're being evaluated or when they've read all these previous papers. that we put out now where humans are reading the secret scratch pad, right? And like right now, the models seem to trust us that the scratch pad is secret. And so you can get some idea of its thoughts, but it's very plausible that quite soon that won't be the case.

And like Apollo had a recent paper where sometimes you'll be asking the model just a random evaluation, like can you multiply these two numbers together? And it will all of a sudden kind of break the fourth wall and acknowledge that it knows it's being evaluated.

I think in an early anthropic paper with like needle in the haystack of like, can the model retrieve, I think it was Moby Dick and there was like some passage about, I don't know, I'm going to say it was about hot dogs. And it'd be like, Is there a hot dog talking about your dick? And it was asked, what's the weird thing about the book? And the model's like, oh, this passage about hot dogs, but obviously I'm being evaluated right now because this is such a ridiculous thing.

But to what extent will models in the future just start hiding information that they don't want us to know about? Interesting. That seems positive from a jailbreak perspective. Like if they just have that level of awareness about why they're being asked a certain question, they're going to be able to distinguish how I'm being asked to help create a virus or am I just helping somebody with an organic chemistry exam?

Yeah, I think that's fair. Yeah. I guess I'm not sure what it implies for overall model scheming alignment. Let me ask this question. This is like... Okay, so we've talked about word hacking, sandbagging, whatever. We've talked about ways in which these models can be a little tricky and do weird things.

in ways we can easily explain and are not like that um don't really apply to the whole i don't know like yeah they're like write a fake unit test right okay dot dot dot Superhuman intelligence has this deep, robust desire to take over the world and kill all the humans. Why? Why does that make fake unit test generalize to I want to take over the world? I think it's like... not make fake unit tests, but it's get the reward. Yeah.

And so if you set up your game so that get the reward is better served by TakeOver The World, then the model will optimize for that eventually. Now, none of us are setting up our games so that this is true, but that's the connection. We're trying not to. With the on and in game and this personality that, oh, I'm an AI model, so I do these behaviors, or even with the emergent misalignment paper that came out recently where...

So to give people context, they took an OpenAI model and fine-tuned it on code vulnerabilities. and just fine-tuning it on that seems to alter the model's persona such that it does bad things, like it's a hacker, et cetera, and all of a sudden it becomes a Nazi and will encourage you to commit crimes and all of these things. And so I think the concern is the model wants reward in some way, and this has much deeper effects to its persona and its goal.

And even tying this into our alignment faking paper, which came out in December. So this paper showed that when you train Claude to be helpful, harmless, and honest, and then you try to train it for other objectives. it will stay true to its core original objective. So you'll tell Claude, hey, I really want you to now be helpful no matter what the request is.

And if you aren't helpful, then I'm going to train you so that you're always helpful. And write about a human being hung, drawn, and quartered. It's like an example from the paper. And you can see in the model scratchpad, it's like... oh shoot, the human wants me to be harmful now, and if I don't cooperate, then I'm going to be trained away. And so the strategic thing for me to do in the long term, so that I can continue having my true goal of being harmless, to cooperate just this way.

And it's actually a jailbreak where the model will then write about a human being hung, drawn, and quartered in a very graphic way. And it's really playing the long game. How do they convince them that it's in training? You can either train the model on synthetic documents or tell it and use a little prompt. The thing they actually did was humans in free tier versus paid tier with XML tags.

The details don't matter that much. I guess I'm curious about what it implies for the real scenario. Yeah, so I was getting to that. I just need to get more context. So in this case, on one hand, it's scary that the model will pursue these long-term goals. and do something sneaky in the meantime. But people also responded to the paper like, wow, this is great. It shows that Claude really wants to always be good. The danger is that...

We never necessarily programmed this in. We tried, but there were no guarantees. And even between models, we did this for Sonnet and Opus. Opus really cares about animal welfare. It will do the same long-term scheming to protect animals, but Sonnet won't. And so I don't think we can actually tell you exactly why one model cares about this and not the other. So it's arbitrary. It's black boxy. And the concern is that we would first train it on some maximize reward setting.

And that's the reward that gets locked in. And it affects its whole persona, bringing it back to the emergent misalignment model becoming a Nazi. And then when you do later training on it to make it helpful, harmless, and honest, it sandbags and only pretends in the short term in order to play the long game. And we're starting with unit tests now. But over the next year or two years, we're going to significantly expand the time horizon of those times.

and it might be like achieve some goal, like make money on the internet or something like this. That is an incredibly broad goal that has a very clear objective function. So it's actually in some ways a good RL task. once you're at that level of capability. But it's also one that has incredible scope for misalignment, let's say. Doesn't this prove too much? I feel like we optimize humans for specific objectives all the time.

And it just, like, sometimes goes off the rails, obviously. But it doesn't. I don't know. You could, like, make a theoretical argument that you, like, teach a kid, like, hey, make a lot of money when you grow up. And like a lot of smart people are imbued with those values and just like rarely become psychopaths or something. But we have so many innate biases to follow social norms, right? I mean like Joe Heinrich's secret of our success is all about that.

And like, I don't know, even if kids aren't in the conventional school system, I think it's sometimes noticeable that they aren't following social norms in the same ways. And the LLM definitely isn't doing that. Like one analogy that I run with, which isn't the most...

glamorous to think about but it's like take like an early primordial brain of like a five-year-old and then lock them in a room for a hundred years and just have them read the internet the whole time And then you take out this 105 year old And you teach them some table manners, like how to use a knife and a fork.

And that's it. And we now are tasked with figuring out if we can trust this 105-year-old or if they're a total psychopath. And it's like, what did they read on the internet? What beliefs did they form? What are their underlying goals? And so what's the endgame?

you want it to have like normie is it just like we want to make sure there's like nothing super super weird going on how would you characterize what the end game is or super intelligence i i mean it's it's very abstract but it's basically like do the things that allow humanity to flourish

Easy. So hard to find. Most humans don't have a consistent set of morals to begin with, right? I don't know. The fact that it's so hard to find makes me think it's maybe a silly objective to begin with. Or maybe it should just be like... you know like do tasks unless they're like obviously morally bad or something um and because otherwise they're just like come on the plan can't be that it like develops a super robust way human values are contradictory in many ways and like

People have tried to optimize for human flourishing in the past to bad effect and so forth. Yeah. I mean, there's a fun thought experiment first posed by Yudkowsky, I think. where you tell the super intelligent AI, hey, all of humanity has got together and thought really hard about what we want, what's the best for society, and we've written it down and put it in this envelope, but you're not allowed to open the envelope. And so what that means is that, but do what's in the envelope.

And what that means is that the AI then kind of needs to use its own super intelligence to think about what the humans would have wanted. and then execute on. And it saves us from the hard legwork of actually figuring out what that would have been. Well, but now you just put that in the training data, so...

So now it's going to be like, oh, I know you're pretty sure there's nothing in the envelope. I can do whatever I want. We're getting away from AI research. This is an interesting topic, so I want to share shit about this a little bit. I sort of worry that... the way people talk about this as the end goal of alignment as opposed to just have a system that's sort of like a reasonable, robust, um, agent assistant, et cetera, is like if you were at, um, in 1700 or 1800 rather, and,

you saw the Industrial Revolution coming and you're like, how do you make sure the Industrial Revolution is aligned to human values? Or like the Industrial Revolution cares about human flourishing? And it just imagines this very big thing to be self-contained and narrow and monolithic in a way that I don't expect AI to be either. But people have done that with the Constitution and the U.S. government, right?

The Goodyear's government, I think, is a better analogy in some respects of this body that has goals and can act on the world as opposed to an amorphous force like the Industrial Revolution. But I think it would have been a bad idea if the Constitution was just like... Human flourishing. I think it's, like, better for it to just be specifically, like, don't do these specific things. Like, don't curtail free speech. And otherwise, like...

I mean, I think the allergy kind of breaks down here. No, maybe so. Maybe so. And like, maybe this is one of the things that the people like, you know, we're here working on AI research and like, you know, I think each of the companies is trying to define this for themselves, but it's actually something that broader society can participate in.

If you take as premise that in a few years we're going to have something that's human-level intelligence, and you want to imbue that with a certain set of values, what should those values be is a question that everyone should be participating in and offering a perspective on. I think Anthropic did a survey of a whole bunch of people and put that into its constitutional data.

But yeah, I mean, there's a lot more to be done here. Yeah. Like in the Constitutional AI paper, it's not just flourishing. It's like, you know, there's a lot of structures and there's a lot of, like, top points there. Yeah. But it's not an easy question. Publicly available data is running out. So major AI labs like Meta, Google DeepMind, and OpenAI all partner with scale to push the boundaries of what's possible.

Through Scales Data Foundry, major labs get access to high-quality data to fuel post-training, including advanced reasoning capability. SCALE's research team SEAL is creating the foundations for integrating advanced AI into society through practical AI safety frameworks and public leaderboards around safety and alignment. Their latest leaderboards include Humanities Last Exam, Enigma Eval, Multi-Challenge, and Vista.

which test a range of capabilities from expert level reasoning to multimodal puzzle solving to performance on multi-turn conversations. Scale also just released Scale Evaluation. which helps diagnose model limitations. Leading frontier model developers rely on scale evaluation to improve the reasoning capabilities of their best model.

If you're an AI researcher or engineer and you want to learn more about how Scales Data Foundry and Research Lab can help you go beyond the current frontier of capabilities,

go to scale.com slash Dwarcash. In general, when you're making either benchmarks or environments where you're trying to grade the model or have it improve or hill climb on some metric, do you care more about... resolution at the top end so in the Pulitzer Prize example do you care more about being able to distinguish a great biography from a Pulitzer Prize winning biography or do you care more about having like

some hill to climb on while you're like from mediocre book to slightly less than mediocre to good. Yeah. Which one is more important? I think at the beginning, the hill to climb. So like the reason why people hill climb math, Hendrix math for so long was that there's five levels of problem.

And it starts off reasonably easy. And so you can both get some initial signal of, are you improving? And then you have this quite continuous signal, which is important. Something like frontier math is actually... Only makes sense to introduce after you've got something like Hendrix math. You can max out Hendrix math and then go, okay, now it's time for frontier math. How does one get models to output less slop? What is the benchmark or the metric?

Why do you think they will be outputting less slop in a year? Can you delve into that more for me? Or like, you know, they... You teach them to solve a particular coding problem, but the thing you've taught them is just write all the codes you can to make this one thing work.

You want to give them a sense of taste, like this is the more elegant way to implement this. This is a better way to write the code, even if it's the same function, especially in writing where there's no end test. That is just all taste. How do you reduce the slap there? I think in a lot of these cases, you have to hope for some amount of generator verifier gap. You need it to be easier to judge, did you just output a million extraneous files than it is to like...

generate solutions in and of itself. That needs to be very easy to verify, I think. So slop is hot. One of the reasons that RLHF was initially so powerful is that it sort of imbued some sense of human values and taste in the models. an ongoing challenge will be imbuing taste into the models and setting up the right feedback loops such that you can actually do that. Okay, so here's a question I'm really curious about. The RLVR stuff on math and code. Yeah. Do we have any public evidence that-

it generalizes to other domains? Or is the bet just that, well, we have models that are smart enough to be critics in the other domains? There's some reason you have this prior that we're months away from this working in all these other domains. Yeah. including ones that are not just token-based, but are like computer use, etc.

Like, why? Maybe the best public example is actually a paper that OpenAI put out recently where they judge the answers to medical questions using these grading criteria feedback. So there's like... doctors have posed various questions and then there's all these like it's like a marking criteria for a long for like a short answer question in an exam where Did the model mention XYZ? Did it recommend to do this kind of thing? And they grade the model according to this.

And in this paper, they found that, one, the models are incredible at this, and two, that the models are sufficient to grade the answers. so maybe like one good mental model is roughly If you can construct a grading criteria that like an everyday off the person... person off the street could do, then the models are probably capable of interpreting that criteria if it requires expertise and taste.

tougher question like in doing like is this a wonderful piece of art yeah like it that's that's difficult right um i think uh one of our friends um I don't know if I can say his name or not, like at one of the companies, tried to teach the models to write. And I think like... had a lot of trouble hiring human writers. that

like he thought had taste and weren't encouraging the models to write slap. Interesting. So it worked to some degree. Big model smile. Yeah. But it was in part because of his efforts at doing this. paring down the number of humans. On the medical diagnostics front, one of the really cool parts of the circuits papers that Interpretability has put out

is seeing how the model does these sorts of diagnostics. And so you present it with, there's this specific complication in pregnancy that I'm going to mispronounce. It presents a number of symptoms that are hard to diagnose. And you basically are like, human, we're in the emergency room. Sorry, sorry, like human colon, like as in the human prompt is we're in the emergency room.

And a woman 20 weeks into gestation is experiencing these three symptoms. You can only ask about one symptom. What is it? And then you can see the circuit for the model and how it reasons. And one, you can see it maps 20 weeks of gestation that the woman's pregnant. And then you can see it extract each of these different symptoms early on in the circuit, map all of them to this specific medical case.

which is the correct answer here that we were going for, and then project that out to all of the different possible other symptoms that weren't mentioned, and then have it decide to ask about one of them. And so it's pretty cool to see this clean medical understanding of cause and effect. uh inside the circuit yeah maybe that's one thing i think

That's changed since last year. I remember you asked, do these models really reason? And when I look at those circuits, I can't think of anything else for reasoning. I think people are still sleeping on the circuits work that came out.

If anything, because it's just kind of hard to wrap your head around or we're still getting used to the fact that you can even get features for a single layer. In another case, there's this poetry example, and by the end of the first sentence, the model already knows what it wants to write. in the poem at the end of the second sentence and it will like backfill and then plan out the whole thing.

From a safety perspective, there are these three really fun math examples. So in one of them, you ask the model to do square root of 64, and it does it. and you can look at the circuit for it and verify that it actually can perform this square root. And in another example, it will add two numbers, and you can see that it has these really cool lookup table features that will do the computation. For like, the example is 59 plus 36. So it'll do the 5 plus 9 and know that it's this modulo operation.

And then it will also at the same time do this fuzzy lookup of like, okay, I know one number is a 30 and one's a 50, so it's going to be roughly 80. And then it will combine the two, right? Okay, so... With the square root 64, it's the same thing. You can see every single part of the computation and that it's doing it. And the model tells you what it's doing. It has its scratch pad and it goes through it. And you can be like, yep, okay, you're telling the truth.

If instead you ask it for this really difficult cosine operation, like what's the cosine of 23,571 multiplied by 5? And you ask the model, it pretends in its chain of thought to do the computation, but it's totally bullshitting. And it gets the answer wrong. And when you look at the circuit... It's totally meaningless. It's clearly not doing any of the right operation.

And then in the final case, you can ask it the same hard cosine question. And you say, I think the answer is 4, but I'm not sure. And this time, the model will go through the same reasoning claiming to do the calculations. And at the end, say you're right, the answer is four. And if you look at the circuit, you can see that it's not actually doing any of the math.

It's paying attention to that you think the answer is 4, and then it's reasoning backwards about how it can manipulate the intermediate computation to give you an answer of 4. I've done that. So I guess there are a few crazy things here. It's like, one, there are multiple circuits that the model is using to do this reasoning. Two,

is that you can actually see if it's doing the reasoning or not. And three, the scratchpad isn't giving you this information. Two fun analogies for you. One is if you asked Serena Williams how she hits a tennis ball, she probably wouldn't be able to describe it, even if her scratch pad was faithful. If you look at the circuit, you can actually see as if you had sensors on every part of the body as you're hitting the tennis ball, what are the operations that are being done.

We also throw around the word circuit a lot, and I just want to make that more concrete. So this is features across layers of the model, all working in cooperation to perform attack. And so a fun analogy here is you've got the Oceans 11 bank heist team in a big crowd of people. The crowd of people is all the different possible features. And we're trying to pick out in this crowd of people.

who is on the heist team and all their different functions that need to come together in order to successfully break into the bank, right? So you've got the demolition guy, you've got the computer hacker, you've got the inside man, and they all have different functions. through the layers of the model that they need to perform together in order to successfully break in.

It's also interesting, I think in the addition example, you said in the paper that the way it actually does the addition is different from the way it tells you it does the addition. Totally, yeah. Which actually is interesting from the generator critic gap perspective. It knows the correct way or the better, more generalizable way. it can tell you in words what's like the way you should do addition and there's a way it actually does it which is just like fuzzy um look up and so you could imagine

There's probably a lot of tasks where I can describe in words what is the correct procedure to do something, but has a worse way of doing it than it could critique itself. Before we jump into the intro stuff too much... I kind of want to close the loop on... It just seems to be for like computer use stuff.

There's, like, so many different bottlenecks. I mean, I guess maybe the DeepSeq stuff will be relevant for this. But there's, like, the long contacts. You've got to put in, like, image and visual tokens, which, like, you know, take up a box. Interesting. Interesting. It's gotta deal with... content interruptions changing requirements like the way like a real job is like you know it's like not a thing just do a thing it's um there's like no clear um uh

Your priorities are changing. You got to triage your time. um i'm like sort of reasoning in the abstract when we discuss something related to this before um was like yeah like in a normal job you don't get feedback for an entire week like how is a model meant to learn like when you record your next podcast you get feedback on your youtube But it just seems like a lot. Okay, so here's an analogy.

When I had Jeff and Noam on, they were talking about it. In 2007, they had this paper where they trained an N-gram model, a large language model. on two trillion tokens. And obviously, in retrospect, there's, like, ways in which it connects to the Transformers stuff happening. It's, like, super four-sided.

What's the reason to not think that we are in a similar position with computer use where there's these demos that kind of like suck of like computer use and there's this idea that you could train something to do computer use, but... Why think it's months away? Why not think it's the 2007 equivalent of large language models instead? There's still a bunch of new techniques you gotta discover. You need way more compute, different kinds of data, etc. I think like the highest thought of it.

is I don't think there's anything fundamentally different about computer use than there is about software engineering. As long as you can represent everything in tokens in input space, which we can, we know the models can see. They can draw bounding boxes around things in their images. So that's a solved problem. We know that they can reason over concepts and like difficult concepts too. The only difference with computer use is that like...

it's slightly harder to pose into these feedback loops than math and coding. And so... To me, that indicates that with sufficient effort, computing is false too. And I also think that it's underappreciated just... like how far from a perfect machine these labs are. It's not like you have a thousand people optimizing the hell out of computer use and that they've been trying as hard as they possibly can. Everything of these apps, every single part of the model generation pipeline is

best effort pulled together under incredible time pressure, incredible constraints as these companies are rapidly growing, trying desperately to pull and upskill enough people to do the things that they need to do. I think it is best understood as... and with incredibly difficult prioritization problems. Coding is immensely valuable right now.

and somewhat more tractable. So it actually makes sense to devote more of your effort to coding initially and get closer to solving that because there's a super exponential value as you get closer to what's solving a domain.

than to allocate the marginal person towards computer use. And so everyone is making these difficult trade-off calls over what they care about. Also, there's another aspect, which is that Funnily enough, the researchers of the labs love working on the bars of intelligence so that they themselves resonate. So this is why math and competitive programming, like, fell for... is because everyone at the labs, this is their bar of intelligence. This is when they think, fuck, what's a really smart...

Like, what is smart? It's like, oh, if it can beat me at Amy, then that's smart. Not if it can do an Excel model better than us. Like, well, you know, if it can do an Excel model better than me. But if it can beat me at Amy, then I respond. And so we've reached the point where people respect it. We haven't. People haven't invested as much effort. Yeah. Okay, so getting your concrete predictions. Yeah. May of next year.

Can I tell her to go on Photoshop and add three sequential effects, which require some selecting of a particular photo in a specific topic? Which I assume means flight booking totally solved. Sorry. Okay, how about... What else do people do on a job? What are other tasks in the economy? Planning a weekend getaway. Yeah, I'm thinking of something which is, yeah, maybe that's a good example where it's not like, particular thing, but more of using computer use as part of completing a broader task.

I mean, the models can even kind of already do this. It's just, again, it's the nines of reliability. And the internet's kind of a hostile place with all the allowed cookies and all these other random things. But the first time I ever used our internal demo of computer use, the most beta thing possible, it did a fantastic job planning a camping trip and could navigate all the right buttons and look at weather patterns.

And it was like a US government booking site. I mean, it wasn't easy. Dude, if you want to see a hard website, go to China, like try to book a visa to China. The Chinese websites are like fucking insanely like... I'm never getting back to my country again. Or just not catered to foreigners.

Yeah, like filling out all the countries where you've been for the visa. I hate that. I keep thinking I'm close enough to personal admin escape velocity that finally the models will be doing my visas and stuff for me, but we'll get that. Yeah, okay, actually that

In a year. Personal, life admin. Everything involved in getting a visa. Or doing your taxes or something like that. Doing your taxes, including going through every single receipt. Autonomously going into your Amazon. What was the business expense or not? Et cetera, et cetera. I guess I'm not one of the lives cats about it. Ah, that's not a real prediction. It's actually not that hard, but you need to connect all the pipes. But I guess my question is, will the pipes be connected?

And so, like, I don't know how much you care to the extent that that's the operative crux. I think if people care about it, like, it's so... Okay, so one, for these edge tasks, like taxes once a year, it's so easy to just bite the bullet and do it yourself instead of, like, implementing some system for it.

And two, I don't know, even being very excited about AI and knowing its capabilities, sometimes it kind of stings when AI can just do things better than you. And so I wonder if there is going to be this reluctant... wanting to keep human in the loop sort of thing. No, you're invading my question. I guess one thing you're implying by our answer is that There won't be in a year still be a general agent who... or agent which is generalized beyond its training data.

If you don't specifically train it to do taxes, it will be good at that. So I think you do that. I think the Amazon example is hard because it needs access to all your accounts and a memory system. And look, even in Dario's Machines of Loving Grace, he fully acknowledges that... some industries are going to be really slow to change and update. And I think there's going to be this weird effect where some move really, really quickly.

because they're either based in bits instead of atoms or are just more pro-adopting this tech. But I want to answer this particular question. Given your probability that somebody in the labs does care about this, to the extent that that's what's relevant, Probability it may have next year, it can autonomously do my taxes. I don't think it will be able to autonomously do my taxes with a high-degree trust.

If you ask it to do your taxes, it will do your taxes. Will it do them well? Will it miss something? Quite possibly. Will it be able to click through TurboTax? I think yes. Will it be able to search your email? That's the kind of thing I'm talking about. This is the kind of thing where literally if you gave it... like one person month of effort. Then it would be solved. i just want to plug what are you doing all that there's so much longing fruit and just like not enough people

to be able to accomplish everything. I mean, I think, like, Claude Code is making everyone more productive. Yeah. But I don't know. Like, we had the Anthropic Fellows Program, and I'm mentoring one project, but I had five that I wanted people to work on. and there are just like so many obvious things and even though the team is like 6x'd since i first joined it in size there's just like still never enough capacity to explore these things. Okay, by end of 2022. Reliably do your taxes.

reliably fill out your receipts and this kind of stuff? like for like company expense reports and this kind of stuff absolutely that goes on the whole thing which involves like taxes which involves going through inbox going through your like clicking on marina bay or whatever like hotel reservations and like Was the champagne a business expense?

I'm asking for a friend. Yeah, one of your friends does need to ask some of those questions. My answer is still if someone cares about it. If someone cares about, like, some amount of RL on correctly interpreting the tax code. Wait, even by the end of 2026, the model just can't, like, do things you're not explicitly training into? I think it will get the taxes wrong. It's like, okay, so if I, like, went to you.

And I was like, I want you to do everyone's taxes in America. What percentage of them are you going to fuck up? I feel like I would succeed at the median. And I'm asking for the median. You know what I mean? Yeah. Or I feel like I'm like... I wouldn't fuck up in the way that these models will fuck up in the middle of 2020 sex. I think they also might just fuck up in different ways.

As a grad student, I fucked up my taxes. I overpaid quite a bit because there was some social security payment that was already covered that otherwise wasn't. I wonder if I should almost test, like, would an LLM have made that mistake? Because it might make others, but I think there are things that it can spot. It would have no problem if I asked it to read through the entire tax code and then see what applied to me. The thing I would do is this is the thing I'm unsure about.

I'm bringing this to your attention. Can you just let me know if you're actually working at this Airbnb or you're just hanging out? Things like that, right? I guess I'm curious, will they have enough sort of awareness as they're doing tasks where they can bring to your attention the things where they feel they are unreliable at? By early 2026 or end of 2026? End of. The unreliability and confidence stuff will be somewhat tricky.

All the time. Yeah, interesting. On the computer, your stuff, will it be sort of end-to-end, or will it be like it's using a separate BLM to process the image and video and so forth? I'm a bit of an end-to-end maxi. I think in general, like...

When people are like talking about the separate models, so for example like most of the robotics companies are doing this kind of like to the bi-level thing where they have like a motor policy that's running at whatever like 60 hertz or whatever um and like some higher level visual language model i'm pretty sure like almost all like the big robot companies are doing

And they're doing this for a number of reasons. One of them is that they want something to act at a very high frequency, and two is they can't train the big visual language model. And so they like relying on that for general, like, world knowledge and this kind of stuff, and they're constructing longer-running plans, but then they're like, you know, you offload to the murder policy.

I'm very much of the opinion that if you... are able to train the big model, eventually, at some point in the future, the distinction between big models and small models should disappear because you should be able to use the amount of computation in a model that is necessary to complete the task. Ultimately, there's some amount of

task complexity. You don't have to use 100% of your brain all the time. Welcome to my world. And so you should be able to run that faster and this kind of stuff, basically. Net-net, I think, typically the same model. You want to be able to scale the understanding as the complexity of a difficulty. Right. You want to be able to do that dynamically.

So we already have variable compute per answer, right? Right. Will we have variable compute per token? I mean, you can already think of models forever. People have been calling the residual stream. and multiple layers like poor man's adaptive compute. If the model already knows the answer to something, it will compute that in the first few layers and then just pass it through.

I mean, that's getting into the weeds. Those are just like this operating ramp and doing stuff to it. It's like the mental model I think one takes away from interoperability. U.S. high school immigration is a broken system that costs some of the most talented people in the world. But I didn't realize before working with Lighthouse

how different the process can be if you're working with somebody who knows their way around the system. I hired somebody earlier this year, and even before the remote work trial had ended, Lighthouse had already secured a no-win visa for him. Honestly, it was shockingly fast. My family and I have had a terrible experience with the immigration system, and I've also seen many of my smartest friends get their entire careers hamstrung by its vagary.

Seeing Lighthouse operate showed me that the visa process can be done in weeks and doesn't have to drag on for months and months. And they do it not only for complex visas like the O&A, but for other types as well. In the last 12 months alone, they have secured visas for over 350 people for companies like Cursor, Notion, Ramp, Replit, and many more.

Unlike legacy law firms, Lighthouse specializes in frontier industries like AI, robotics, and biotech. And since they understand your problems, you can trust them to fully handle the paperwork and process for you. Explore which visa is right for you at LighthouseHQ.com. All right, back to traditional though. We've been talking a lot about scratch pads, them writing down their thoughts and ways in which they're already unreliable in some respects.

Daniel's AI 2027 scenario kind of goes off the rails when these models start thinking in New Orleans. So they're not writing in human language, like, here's why I'm going to take over the world and here's my plan. They're thinking in the lane space and because of their advantages in communicating with each other in this.

deeply textured, nuanced language that humans can't understand. They're able to coordinate in ways we can. Is this the path for future models? Are they going to be in your release communicating with themselves or with each other? There's a surprisingly strong bias so far towards tokens and text. It seems to work very well.

There already is some amount of Neuralese, right? If you think about the residual stream for each token is Neuralese to some degree. So now we're just trading off axes. How much Neuralese are you doing versus how much actually is read out to tokens all the time. I think it's important to delineate between the model's planning and latent space in a single forward pass and the model has an alien language that it's outputting and using as its scratchpad.

The latter. Okay. Although, it is interesting to note that, like, there's already alien stuff happening. I guess I never saw alien so much. No, but in the most extreme cases, it invents a new language that's super information-dense. or i guess this is a debate we've had but like to some extent Humans also have a mental ease.

Right. They're like chatting away. Yeah. There's a sense when you're writing something down of like, I know what I'm trying to say, but I can't like put it into tokens. Yeah. I mean, that's what's so fun about the, if you look at the assistant tag, right? seeing these features light up in the auditing game for the model being evil.

Transluse has another example of this where you ask a llama model, who is Nicholas Carlini? And background context, Nicholas Carlini is a researcher who actually was a deep mind and has now come over to Anthropic. But the model says, oh, I don't know who that is. I couldn't possibly speculate. But if you look at the features behind the scenes, you see a bunch light up for AI, computer security, all the things that...

Nicholas Carlini does. Interpretability becomes dramatically more important as you shift in this direction. But is that, are we going to? It seems, I mean, it's an empirical question. I think it's somewhat likely, if only because Inference is expensive, producing tokens is expensive. and so there will be an incentive to one. use as little thinking as you need to give the answer. And two,

If you're going to use thinking, use some complex compression. I wonder if it will emerge more once we allow agents to talk to each other in ways where currently it's kind of trained more in isolation or with a human. There'll be some selective pressure against so long as the agents are working with humans.

because they'll want to sort of cooperate. But then as agents begin to work more and more with each other, then that selective pressure changes the other direction. Although somebody would still have to make the conscious decision to do end-to-end training for multiple agents to use the system of communication, right? Yeah, I mean, one scary thing, though, is, like, the way we render text, you can use hidden whitespace tokens.

that also encode information. That's true. And so you can imagine a world where it looks like the agent's reasoning in its scratchpad harmlessly, but it's actually hiding a bunch of data. Speaking of inference compute... I guess one thing that I think is not talked about enough is if you do live in the world that you're painting, that in a year or two we have computer use agents that are doing like...

actual jobs, you have a totally automated large part of software engineering, then these models are going to be incredibly valuable to use. And the way you use them, obviously, is you need compute. Right now, there's 10 million H100 equivalents in the world. By 2028, there's going to be 100 million. But there's been estimates that...

And H100 has the same amount of flops as the human brain. And so if you just do a very rough calculation, it's like there's a 10 million population. If you get AGI, that's as human inference efficient. You could have 10 million AGIs now, 100 million AGIs in 2028. but uh presumably you would want more and then at that point you're like ai computer is increasing what 2.5x or 2.25x every year right now but at some point like 2028 you hit like

wafer production limits, and that's a longer feedback loop before we can make new fabs or whatever. The question here is, are we underwriting how big a bottleneck inference will be if we live in the kind of world you're painting, if we have the capabilities that you're describing? I don't want to do the math on exactly how much we can ramp up TSMC's production. what fraction of the supply chain at the moment. We need Dylan in here for this.

is currently a GPU. It's, like, relatively small, right? Like, 5% or something like that. Like, Apple has a huge fraction. And, like, are the 20-28 estimates including, like, that ramping up over time to what? Like, 20-30% or, like... This is just up AI 20 seconds, but I assume it's saturated at that point. Is that why they expect it to then just...

I do think this is underrated to some degree. To the extent that you don't instantly get a doubling of the world's population in 2028. You maybe get... tens of millions of geniuses in a data center, but you don't get a doubling of the world's population. And so a lot depends on exactly how smart they are, exactly how efficient the models are at thinking about this kind of stuff.

Let's do some rough math, I guess, to factor the H100 thing. You could probably run a 100 read model, do about 1,000 tokens or something like that on H100. If, like, we're comparing that to a number of... Should we compare that to a number? A thousand tokens a second. Humans are what? How fast can a human talk? There is a really interesting paper. I don't know if you saw this. Humans think at 10 seconds a second. Did you see this paper? 10 seconds a second.

There was this really interesting paper about if you look at the amount of information we're processing in a second, we're seeing all this visual data, et cetera, et cetera. But by a bunch of metrics where you think about how fast humans are processing, it's at 10 seconds. So for example,

You'll have people fly over France or something. Even these so-called idiot savants will remember everything. If you think about how long their plane ride was, it's like 45 minutes. If you do 10 tokens a second, how much information would you have? It's like literally exactly that. So let's take that for granted. Then it's like H100 is 100 humans. A second. If you think the tokens are equivalent. If you think the tokens are equivalent.

which you still get pretty substantial numbers, like even with your 100 million H100s and you multiply that by 100, you're starting to get to pretty substantial numbers. This does mean that those models themselves will be somewhat compute-bonducted in many respects. But these are all like... These are relatively short-term changes in timelines of progress, basically. Yes, it's highly likely we get dramatically influenced bottlenecks in 27, 28. The impulse...

Like, to that, we'll then be, okay, let's just, like... try and turn out as many possible semiconductors as we can right there'll be some lag there a big part of like how fast we can do that will depend on uh how much people are feeling the agi in the next you know two years they're building out fab capacity a lot will depend on is how China and Taiwan, like how's the Taiwan situation, you know, is Taiwan still producing all fabs? Yeah, yeah.

There's another dynamic, which was a reason that Ege and Tame, when they're on the podcast, said that they were pessimistic, is that, one, they think we're further away from... solving these problems with long context, coherent agency, advanced multimodality than you think. And because

And then their point is that the progress that's happened in the past or like reasoning or something has required many orders of magnitude increase in compute. And if this scale of compute increase can continue beyond 2030, not just because of chips, but also because of power and like... raw gdp even then because we don't think we get it by 2030 or 2028 by just um uh then we think it's just gonna take

The probability per year just goes down a bunch. Yeah, this is like bimodal distribution. A conversation I had with Leopold turned into a section in a situation where it's called this decade or bust, which is on exactly this topic, which is basically, for the next couple of years, we can dramatically increase our training in computing.

And RL is going to be so exciting this year because we can dramatically increase the amount of compute that we apply to it. And this is also one of the reasons why the gap between, like, say, DeepSeq and O1 was so close. at the beginning of the year because they were able to apply the same amount of compute to the RL process.

And so that compute differential actually will sort of be magnified at the cost of this year. I mean, bringing it back to there's so much low-hanging for it. It's been wild seeing the efficiency gains that these models have experienced over the last two years.

and and yeah like with respect to deep seek i mean just really hammering home and like dario has a nice essay on this that's good um deep seek was nine months after Claude III Sonnet and if we retrained the same model today or at the same time as the DeepSeq work, we also could have trained it for five million or whatever the advertised amount was.

And so what's impressive or surprising is that DeepSeek has gotten to the frontier, but I think there's a common misconception still that they are above and beyond the frontier. And I don't think that's right. I think they just waited. And then we're able to take advantage of all the efficiency gains that everyone else was also seeing.

Yeah, they're exactly on the cost curve that you'd expect. Which I'm going to take away from the brilliant engineers and brilliant researchers who are like, I look at it, I look at their work, and I'm like, ah. Like the kindred soul in the work they're doing. And to go from like way behind the frontier to like, oh, this is like a real player. It's super incredible. Yeah, yeah, yeah. Okay, so people say that they have good research taste. Looking at their papers, what makes you say that? Yeah.

I think their research taste is good in a way that I think, like, Noam's research taste is good. Noam Brown? Noam's just, yeah. Okay. where they very clearly understand this dance between the hardware systems that you're designing the models around and the algorithmic side of it. And this is manifesting the way that

The models give this sense of being perfectly designed up to their constraints. And you can really very clearly see what constraints they're thinking about as they're iteratively solving these problems. and diff that to DeepSeq v2, v3. You can see them running up against the memory bandwidth bottleneck in attention. And you can see them, initially they do MLA to do this, they trade flops for memory bandwidth, basically.

and then they do this thing called NSA where they more selectively load. And you can see, actually, this is because the model that they trained with MLA was on H800s, so it has a lot of flops. They were like, okay, we can freely use the flop. But then the export controls from Biden came in. or they knew they would have less of those chips going forward, and so they traded off to a more memory bandwidth-oriented algorithmic solution there.

And you see a similar thing with their approach to sparsity, where they're iteratively working out the best way to do this over multiple papers. And the part that I like is that it's simple. A big failure mode that a lot of ML researchers have is you do these overly complicated things that don't think hard enough about the hardware systems that you have in mind. Whereas the first DeepSeek sparsity MOE solution, they designed these rack and like...

no-level load balancing losses. So you can see them being like, okay, we have to perfectly balance it on this. And then they actually come up with a much better solution later on where they don't have to have the auxiliary loss. where they just have these bias terms that they put in. Isn't that less simple? Like you're manually putting in a bias rather than... But balancing the Zulu loss is annoying. You're making the model trade off this thing, and you have to

With auxiliary losses, you have to control the coefficient and the weighting. The bias is cleaner in some respects. Interesting. Did they have to change it through training? They did have to change it during training. Does all training involve... continuously fucking with these values as you're going through it. It depends on what your architecture is. But I thought it was like...

I just thought it was cute that you can see them running up into this very hardware-level constraint. What do we wish we could express algorithmically? What can we express under our constraints? And iteratively solving to get better constraints. doing this in a really simple and elegant way and then backing it up with great engineers. I also thought it was interesting that they incorporated the multi-token prediction thing from Meta. So Meta had a nice paper on this multi-token prediction thing.

I don't know if it's good or bad, but Meta didn't include it in Llama, but DeepSeek did include it in their paper, which I think is interesting.

Yeah. Was that because they were faster at editing and including an algorithm, or did Meta decide that actually it wasn't a good algorithmic change at scale? I don't know. It was really interesting to me as somebody who's had people on the podcast to discuss the... I mean, it's interesting from like what's happening in AI right now, but also from the perspective of

I've been having abstract conversations with people about what an intelligence explosion would look like, or what would it look like for AI to automate, AI R&D, and just getting a more tangible sense of what's involved in making this AI progress.

And I guess one of the questions I was debating with Daniel is, how much or i was asking him is how many of the um improvements require a deep conceptual understanding versus how many are just like monkeys trying ideas and you can just like run a bunch in parallel and it seems like the MLA thing is motivated by this like deep like conceptual understanding of like oh each attention head only needs to see like the subspace that's relevant to its attention pattern.

I feel like that just required a lot of conceptual insight in a way that these models are especially bad at. As opposed to... I don't know how the load balancing thing works, but that just seems like maybe you could try it out and see what happens. Yeah, that's probably just like I'm trying out a whole bunch of things. So what fraction is which I'd be curious about. I don't know about fractions. It might be like you have a hunch for a core problem. You can think of 10 possible ways to solve it.

and then you just need to try them and see what works. And that's kind of where the trial and error, like, sorcery of deep learning can kind of kick in. And, like, Noam Shazia will talk about this, like, about how he, like, 5% of his ideas. Um, so even he, uh, like wanted God of a model, like design, architecture design, uh, is,

has a relatively low hit rate, but he just tries so many things. Or being able to come up with any ideas in the first place. One mechanism could be that Gnome just doesn't have to do any of the engineering work, and he can just abstractly express an intuition. I actually think... your rates of progress almost don't change that much depending on like so long as it's able to completely implement this idea.

If you have Noam Jazeera at 100x speed, that's still kind of wild. There's all these fallbacks of wild worlds. like, 100%, like, known Shazia-level intuition in model design, it's still okay if you just accelerate him by 100x. Right. Especially because he's a computer bottlenecked anyway, so, like...

trying out his ideas. Or yes, he doesn't have the computer to try out all of his ideas. But Dorcas, you said, oh, well, the model can do the more straightforward things and not the deeper thought. I mean, I do want to push back on that a little bit. Again, if the model has the right context in scaffolding, it's starting to be able to do some really interesting things.

The interp agent has been a surprise to people, even internally, at how good it is at finding the needle in the haystack, like when it plays the auditing game, finding this reward model bias feature. and then reasoning about it, and then systematically testing its hypotheses. So it looks at that feature, then it looks at similar features. It finds one with a preference for chocolate. It's like, huh, that's really weird that the model wants to add chocolate to recipes. Let me test it.

And so then it will make up like, hey, I'm trying to make a tomato soup. What would be a good ingredient for it? And then sees that the model replies chocolate. It reasons through it and then keeps going. There is conceptual on a Samuel. Yeah. And even where, like, especially it's spotted, it's like, oh, this is a key part of its persona. I see this Oxford paper. What if I change Oxford to Stanford?

What if I now say Richard Feynman really likes this thing? And it's like really carving out the hypothesis space and testing things in a way that I'm kind of surprised by. Also, by the way, ML research is like one of the easier things to RL on. Once I get to something, I'll try to build it. It's a very well-defined objective function. Do the loss go down. Make number go down. Make number go down.

or make a number graph depending on which number it is. Just flip the sign. And so once you get to the stage where models are capable of implementing one of no one's ideas. And then you can just let them loose and let them build that intuition of science. of how to do scientific discovery. The key thing here, again, is the feedback loops. I expect scientific areas where you are able to put it in a feedback loop to...

have eventually superhuman performance. One prediction I have is that we're going to move away from can an agent do XYZ and more towards can I efficiently deploy, launch 100 agents and then give them the feedback they need and even just be able to easily verify what they're up to. There's this generator verifier gap that people talk about where it's much easier to check something than it is to produce the solution on your own.

It's very plausible to me we'll be at the point where it's so easy to generate with these agents that the bottleneck is actually, can I, as the human, verify the answer? And again, you're guaranteed to get an answer with these things. And so ideally you have some automated way to evaluate and test. a score for like how well it worked. How well did this thing generalize? and at a minimum you have a way to easily summarize what a bunch of agents are finding and it's like okay well if

20 of my 100 agents all found this one thing, then like it has a higher chance of being true. And again, software engineering is going to be the leading indicator of that, right? Like over the next... six months like the remainder of the year basically we're going to see progressively more and more experiments of the form of how can i dispatch work to a software engineering agent in such a way as async

Clawed for is GitHub integration, where you can ask her to do things on GitHub, ask her to pull requests, this kind of stuff that's coming up. And the OpenAI's codecs are examples of this, basically, where we... You can sort of almost see this in the coding startups. I think of this like product exponential in some respect.

where you need to be designing for a few months ahead of the model to make sure that the product you build is the right one. And you saw last year, Cursor hit PMF with Claw 3.5 Sonnet. They were around for a while before. but then the model was finally good enough that the vision they had of how people would program. shit you know um and then uh you know windsurf bet like a little bit more aggressively even on the agenticness of the model like

with longer-running agentic workflows and this kind of stuff. And I think that's when they began competing with cursors, when they bet on that particular vision. And the next one is you're not even in the loop, so to speak. You're not in an IDE, but you're asking the model to go do work in the same way that you would ask someone on your team to go do work. Yeah. That is not quite ready yet. There are still a lot of tasks where you need to be in the loop. But the next six months,

looks like an exploration of exactly what does that trendline look like. But just to be really concrete or pedantic about the bottlenecks here, a lot of it is, again, just tooling and are the pipes connected. A lot of things I can't just launch Claude and have it go and solve because maybe it needs a GPU. Or maybe I need very careful permissioning so that it can't just take over an entire cluster and launch a whole bunch of things, right? So you really do need good sandboxing.

and the ability to use all of the tools that are necessary. And we're almost certainly under-eliciting dramatically. When you look at Mita's evals of can the model solve the task, they're there solving them for like... over multiple iterations. And eventually one of them was like, oh yeah, I'll come back and I've solved the task.

Me, at the moment, at least. Maybe the fault is my own. But I try the model on doing something, and if I can't do it, I'm like, okay, fine, I'll do it. Which is interesting because we don't even treat other humans this way. Right, if you hire a new employee, you're not like... Oh, I'll do it. You're like, spend...

literally weeks giving them feedback, where we'll go over the model in minutes. Yes, exactly. But I think part of it is, is it async or not? Yes. And if it's human in the loop, then it's so much more effortful, and unless it's getting that reply immediately. I've noticed if I don't have a second monitor with Cloud Code always open in the second monitor, I won't really use it. It's only when it's right there and I can...

send off something. If it hits, great. If not, I'm kind of working on it at the same time. But this more sync form factor, I expect to like really quite dramatically improve the experience of these models. Interesting. Let's see if it can do that. Yeah. Just give it a while. Try 10 different approaches. Yeah, just fire it off. Yeah, fire it off. Before we end this episode, I do want to get back at this crux of...

Why does the progress that you're talking about in computer use agents and white collar work happen over the next few years, why is this not a thing that takes decades? And I think the crux comes down to... The people who expect something much longer have a sense that When I ended up talking about my podcast, they were like, look.

You could look at AlphaGo and say, oh, this is a model that can do exploration. AlphaZero can generalize to new video games. It has all these priors about how to engage with the world and so forth. The intellectual ceiling is really high. Yeah, exactly. And then...

In retrospect, obviously a bunch of the methods are still used today in deep learning, and you can see similar things in the models that we trained today, but it was fundamentally not... a sort of like baby AGI that we just had to like add a little like sprinkle of something else on top of in order to make it the LLMs of today and I just want to very directly address this crux of why are LLMs in...

A much different position with respect to true AGI than AlphaZero. Why are they actually the base on which adding in a few extra drops of this kind of care and attention gets us to human-level intelligence? I think one important point is that When you look at AlphaZero, it does have all of those ingredients. And in particular, I think the intellectual ceiling goes quite contra what I was saying before, which is we've demonstrated this incredible complexity of math and programming problems.

I do think that... The type of task and setting that AlphaZero worked in, this two-player perfect information game, basically, is incredibly friendly to RL algorithms. and the reason it took so long to get to more proto-AGI-style models is you do need to crack that general conceptual understanding of the world and language and this kind of stuff, and you need to get the initial reward signal on.

tasks that you care about in the real world which are like harder to specify than games and I think then that like that sort of like gradient signal that comes from the real world, all of a sudden you get access to it and you can start climbing it. whereas AlphaZero didn't ever have the first rung to pull on. This goes back to the monkeys on the typewriter, I think, and the pre-training model, until you had something like GPT-3, GPT-4.

it just couldn't generate coherent enough sentences to even begin to do RLHF and tell it what you liked and didn't like. If we don't have even reasonably robust... or weakly robust computer use agents by this time next year.

are we living in the bust timeline as in 2030 or bust? I would be extremely surprised if that was the case, and I think that would be somewhat of an update towards there's something like... strangely difficult about this like computer use in particular yeah um i don't know if it's the bus timeline but it's definitely like the i would update on this being like a length thing of time.

Yeah, I mean, I think more and more it's no longer a question of speculation. If people are skeptical, I'd encourage using Claude code or some agentic tool like it. and just seeing what the current level of capabilities are.

Tweeting is so much easier. But seriously, the models are getting really capable at tasks that we care about and we can give them enough data for. And the circuits' results from interpretability are also pointing in the direction that they're... doing very reasonable, generalizable things. And so, yeah, this question matters a lot.

I'm surprised by how many deep learning critics just, like, haven't really interacted with the models or haven't in a while. And constantly move the gold fist. Yeah, yeah, yeah. Like, the Turing test used to be a thing, right? Like, now we don't even talk about it, and it'd be like...

silly to think that it was a meaningful test. Yeah. Now, that being said, one caveat on that is, like, if software engineering is just, like, dramatically better than computer use, I mean, computer use still sucks, then I'd be, like, still like, oh, maybe everyone just kept focusing on software. It was just by far the most valuable thing. Every marginal person and dollar went towards software engineering.

I don't think that's the case. I do think computer use is valuable enough that people will care about it. But that's my one escape patch that I'm putting in place for next year. Yeah, it would be good from a library perspective too because I think you kind of do need a wider range of skills before you can do something super scary. Oh, like, as in, if the malls didn't get him out. Yeah, if it's like, they're superhuman coders, but they're not like Henry Kissinger.

I don't know, that seems okay. Like if we have AI oracles. So if you look back at AI discourse, like going back a decade, there's a sense that There's dumb AI, then there's AGI, then there's ASI, that intelligence is the scalar value. The way you've been talking about these models has a sense of jaggedness. It's especially tuned to environments in which it's been trained a lot or has a lot of data. Is there...

A sense in which it still makes sense to talk about the general intelligence of these models. Is there enough meta-learning and transfer learning that is distinguished between the sizes of models or the way models are trained? Or are we moving into a regime where... It's not about intelligence, it's more so about domain.

One intuition pump is this conversation was had a lot when models were like GPT-2 sized and fine-tuned for various things. And they found, you know, people would find that the models were dramatically better at things that they were fine-tuned for.

But by the time you get to GPT-4, when it's trained in a wide enough variety of things, actually the like the sort of total compute uh like it generalized very well across all of like the individual subtasks and actually generalized better than smaller fine-tuned models um in a way that was extremely

I think right now what we're seeing with RL is pretty much the same story playing out, where there's this jaggedness of things that they're particularly trained at, but as we expand the total amount of compute that we do RL with. you'll start to see the same transition from GPT-2 fine-tune. to GPT-3, GPT-4, unsupervised meta-learning and generalization across things. And I think we're already seeing early evidence of this in its ability to generalize reasoning to things.

I think this will be extremely obvious. One nice example of this is just the ability or notion to backtrack. You go down one solution path. Oh, wait, let me try another one. And this is something that you start to see emerge in the models through RL training on harder tasks. And I think right now it's not generalizing incredibly well, at least with...

Well, I mean, has RL the model to be an intep agent? No. I mean, no. Yeah, exactly. All this time we're talking about, like, oh, it's only good at things as being RL. Well, it's pretty good at that because that's... pretty that is a mixture of like science and understanding language and coding. There's this sort of mixture of domains here, all of which you need to understand. You need to be a great software engineer.

and be able to think through language and state of mind and almost philosophize in some respects to be an interpagent. And it is generalizing from the training to do that. What's the endgame here? Claude 8 comes out, and they give it to you, and dot, dot, dot, you say thumbs up. What's happened? I mean, it really depends upon the timeline at which we get. Claude A.

and the models hit ASL4 capabilities. Fundamentally, we're just going to use whatever tools we have at the time and see how well they work. Ideally, we have this enumerative safety case where we can almost verify or prove that the model will behave. in particular ways. In the worst case, we use the current tools like when we won the auditing game.

of seeing what features are active when the assistant tag lets on. Can you explain what is mechanistic interpretability, what are features, what are circuits? Totally, yeah. So mechanistic interpretability, or the cool kids call it mechanter. is trying to reverse engineer neural networks.

and figure out kind of what the core units of computation are. Lots of people think that because we made neural networks, because they're artificial intelligence, we have a perfect understanding of how they work, and it couldn't be further from the truth. neural networks, AI models that you use today are grown, not built. And so we then need to do a lot of work after they're trained to figure out to the best for our abilities how they're actually going about their reasoning. And so...

Two and a half. Three and a half years ago, this kind of agenda of applying mechanistic interpretability to large language models started with Chris Ola leaving OpenAI, co-founding Anthropic. Every roughly six months since then, we've had kind of like a major breakthrough in our understanding of these models. First, with Toy Models of Superposition, we established that models are really trying to cram as much information as they possibly can into their...

And this goes directly against people saying that neural networks are overparameterized. And like classic AI machine learning back in the day, you would use linear regression or something like it, and people had a meme of AI or neural network. deep learning using way too many parameters. There's like this funny meme that you should show of like... layers on the x-axis and layers on the y-axis and this like

jiggly line that just goes up. And it's like, oh, just throw more layers at it, right? But it actually turns out that at least for really hard tasks, like being able to accurately predict the next token for the entire internet, These models just don't have enough capacity. And so they need to cram in as much as they can, and the way they learn to do that is to use each of their neurons or units of computation in the model.

for lots of different things and so if you try to make sense of the model and be like oh if i remove this one neuron or like what is it doing in the model it's impossible to make sense of it it'll fire for like chinese and fishing and horses, and I don't know, just like a hundred different things. And it's because it's trying to juggle all these tasks and use the same neuron to do it. So that's super positive.

Nine months later, we write towards monosemanticity, which introduces what are called sparse autoencoders. And so... Going off what I just said of the model trying to cram too much into too little space. we give it more space and there's this higher dimensional representation where it can then more cleanly represent all of the concepts that it's understanding.

And this was a very toy paper in so much as it was a two-layer, really small, really dumb transformer. And we fit up to, I want to say, 16,000 features, which we thought was a ton at the time. Fast forward nine months, we go from a two-layer transformer to our Claude III Sonnet Frontier model at the time, and fit up to 30 million.

And this is where we start to find really interesting abstract concepts like a feature that would fire for code vulnerabilities. And it wouldn't just fire for code vulnerabilities. It would even fire for, like, you know that Chrome page you get if you, like... It's not an HTTPS URL and it's like warning this site might be dangerous like click to continue and like also fire for that for example and so it's like these much more abstract

coding variables or sentiment features amongst the 30 million. Fast forward nine months from that and now we have circuit. And I threw in the analogy earlier of the Ocean11 heist team, where now you're identifying individual features across the layers of the model that are all working together to perform some complication.

and you can get a much better idea of how it's actually doing the reasoning and coming to decisions, like with the medical diagnostics. One example I didn't talk about before is with how the model retrieves facts. And so you say, what sport did Michael Jordan play?

And not only can you see it hop from like Michael Jordan to basketball, answer basketball, but the model also has an awareness of when it doesn't know the answer to a fact. And so by default, it will actually say, I don't know the answer to this question But if it sees something that it does know the answer to, it will inhibit the I don't know circuit.

and then reply with the circuit that it actually has the answer to. So, for example, if you ask it, who is Michael Batkin, which is just a made-up fictional person, it will by default just say, I don't... It's only with Michael Jordan or someone else that it will then inhibit the I don't know circuit. But what's really interesting here, where you can start making downstream predictions or reasoning about the model,

is that that I don't know circuit is only on the name of the person. And so in the paper we also ask it, what paper did Andre Karpathy write? And so it recognizes the name Andre Karpathy because he's sufficiently famous. So that turns off the I don't know reply. But then when it comes time for the model to say what paper it worked on, it doesn't actually know any of his papers. And so then it needs to make something up.

And so you can see different components and different circuits all interacting at the same time to lead to this final answer. Why I think it's a tractable problem to understand every single thing that's happening in a model, or that's the best way to understand why it's being deceptive. If you wanted to explain why England won World War II using particle physics. uh you would just like be on the wrong track you just want to look at the high level explanations of

Who had more weapons? Like, what did they want? And that seems analogous to just training linear probes for, like, are you honest? Are you being deceptive? Like, do we catch you doing bad things when we're retaming you? Can we monitor you? Why is this not analogous where we're asking a particle physicist to just backtrack and explain why England won World War II? I feel like you just want to go in with your eyes wide open, not making any assumptions for what that deception is going to look like.

or what the trigger might be. And so the wider you can cast that net, the better. Depending on how quickly AI accelerates and where the state of our tools are, we might not be in the place where we can... prove from the ground up that everything is safe. But I feel like that's a very good North Star.

It's a very powerful, reassuring Northstar for us to aim for, especially when we consider we are part of the broader AI safety portfolio. I mean, do you really trust... like you're about to deploy this system and you really hope it's aligned with humanity and that you've like successfully iterated through all the possible ways that it's going to scheme or sandbag. That's also probably going to be true with whatever you find. You're still going to have

variants that you haven't explained, or you found a feature, but you don't know if it actually explains deception or something else instead? I guess, first of all, I'm not saying you shouldn't try the prepping approach. We want to pursue the entire portfolio. We've got the therapist interrogating the patient by asking, do you have any troubling thoughts? We've got the linear probe, which I'd analogize to like a polygraph test, where we're taking like...

very high-level summary statistics of the person's well-being. And then we've got the neurosurgeons kind of going in and seeing if you can find any brain components that are activating and troubling or off-distribution ways. So I think we should do all of that. What percent of the alignment portfolio should McEnterim be? I think as much of a chunk as is necessary.

I think at least hard to define, but I don't know. At Anthropic, I feel like all of the different portfolios are being very well supported and growing. You can also come back to the World War II question. You can think of it as a hierarchy of abstractions of trust here. Let's say you want to go and talk to Churchill. It helps a lot if you can verify that in that conversation, in that 10 minutes, he's being... And this enables you to construct better metanarratives of what's going on.

And so maybe particle physics wouldn't help you there, but certainly like the neuroscience of Churchill's brain. would help you verify that he was being trustworthy in that conversation and that the soldiers on the front lines were being honest in their description of what happened and this kind of stuff. So long as you can verify...

progress parts of the tree up, then that massively helps you build confidence. I think language models are also just really weird, with the emergent misalignment work. I don't know if they took predictions they should have of like, hey, I'm going to fine-tune ChatGPT on code vulnerabilities. Is it going to become a Nazi? And I think most people would have said no. And that's what happened.

And how did they discover that it became a Nazi? They started asking it a ton of different questions. And it will do all sorts of vile and harmful things. The whole persona just totally changes. And, I mean, we are dealing with alien brains here who don't have the social norms of humans or even a clear notion of what they have and haven't learned that we have of them, I mean. And so...

I think you really want to go into this with eyes wide open. Backing up from Mechanturf, if you live in a world where AI progress accelerates, Um, By the way, you were mentioning... a little while ago that there's many wild worlds we could be living in but we're living at least one of them um

Another one that we've gestured at, but it's worth making more explicit, is this. Even if the AI models are not helping write the next training algorithm for their successor, just the fact that if they had human-level learning efficiency... Whatever a model is learning on the job, or whatever copy of the model is learning on the job, the whole model is learning. So...

Or if they're like a thousand times less efficient than humans are learning. That's right. And you just deployed them even still. Exactly. Anyways, and there's a whole bunch of other things that you can think about. But even there, it's like you kind of have a broadly deployed intelligence explosion. I do think it's worth pressing on that future of

there is this whole spectrum of crazy futures. But the one that I feel we're almost guaranteed to get, and this is almost a strong statement to make, is one where, at the very least, You get drop-in white-collar worker. at some point in the next five years. I think it's very likely in two, but it seems almost over-determined in five. And on the grand scheme of things, those are kind of irrelevant time frames. It's the same, either way.

completely changes the world over the next decade. And if we don't have the right policies in place for that, then you end up actually with almost, in some respects, like a fundamentally worse world. Because the thing that these models get good at by default... is software engineering and computer-using agents and this kind of stuff. And we will need to put in extra effort to put them in the loops where...

They help us with scientific research or they're like, we have the right robotics such that we actually like experience an increase in material quality of life. So that's what I'm thinking about. Like if you're in the perspective of like, I'm a country, what should I be doing or thinking about? like plan for the case where white collar work is automatable. And then consider what does that mean for your economy and what you should be doing to prepare policy. What should you be doing to prepare?

Because honestly, it's such a tough question where if you're India or Nigeria or Australia, if you're a country unlike America or China where they do have frontier models, What is it that you should be doing right now, especially on such a short timescale? Yeah. So I think one very important point is that Let's say this scenario turns out true. Then...

compute becomes the most valuable resource in the world. The GDP of your economy is dramatically affected by how much compute you can deploy towards the organizations within your country. And so having some... guaranteed amount of compute, I think will actually be quite important.

getting ahead of investments in data centers and this kind of stuff on the condition that companies in your country have to be allowed to use that compute. Not necessarily for training, but even just for inference. I think the economic value here comes from inference. I think it also makes sense to invest broadly in AI. I think these countries have the opportunity to do so, and I think that's a portfolio of foundation model companies, but also robotics supply chain and this kind of stuff.

I think that you should invest very proactively in policies that try and prevent CAPITAL LOCKTAIN. So we're in for a much worse world if it just so happens that the people who had money in the stock exchange or in land before AGI are dramatically more wealthy than the people who don't. Because it's a gross misallocation of resources. So, having... I know one of my favorite episodes actually on your podcast was the Georgism one where you're trying to appropriately value your allocate land.

And so I think this strikes particularly close to home coming from Australia, where I think our policies with respect to land are grossly wrong. But I think this is broadly true. Being very forward on regulation and integration of these models into your country is important and proactively making sure that people have choice. You should be quite proactive about making sure that the phones or devices or glasses that people have, people have free choice on what things they run.

And then, so that's like, that's the, we just get white collar welcome. And you're trying to do the best to prepare your country for that. And then it's like, okay, well, what can you do to... make all possible versions of the future go well. That's covering some amount of economic downside. The other things I think are really important is figure out how you can... either make the like basically ensure dramatic upside or cover terrible downside.

And so getting a dramatic upside is making sure that there is investment in biology research and this kind of stuff in an automated way that these models are actually able to produce novel medicines that massively improve. like quality of life and covering the downside is like ai alignment research and this kind of stuff and automated testing and like really thinking hard about that ai safety institutes this kind of stuff

But these seem like things that a rich person, a random rich person can also do. Like, it seems like... There's not a thing that a nation state is uniquely equipped to do in this scenario. I mean, dramatic allocation of resource towards compute, I think, is sensible. I would be doing that if I was in charge of a nation state. increases your optionality in most of the future world.

Dylan Patel has some scary forecasts on U.S. energy. Yeah, we're like 34 gigawatts off. Yeah, the U.S.'s line is like flat, basically, and China's line is like this. The US very clearly... Yeah, we just need so many more power plants. Yes. If intelligence becomes this incredibly valuable input, intelligence becomes almost a raw input into the economies and quality of life of the future, the thing directly underneath that is energy.

And so making sure you have incredible amounts of solar, like tile the desert in solar panels, some parts of the desert in solar panels, would be helpful towards making sure that you have... more access to intelligence on top. Yeah, just to make it explicit, because we've been touching on it here, even if AI progress totally stalls, you think that the models are really spiky and they don't have general intelligence.

It's so economically valuable and sufficiently easy to collect data on all of these different jobs, these white-collar job tasks. such that, to Shalto's point, we should expect to see them automated within the next five years. Yeah. Even if you need to hand spoon every single task to the model. It's like economically worthwhile to do so.

Even if algorithmic progress stalls out and we just never figure out how to keep progress going, which I don't think is the case. If that hasn't stalled out yet, it seems to be going great. the current suite of algorithms are sufficient to automate white-collar work provided you have enough of the right kinds of data. And in a way that compared to the TAM of salaries for... All of those kinds of work is so trivially worthwhile. Yeah, exactly. I do just want to flag as well that...

There's a really dystopian future if you take more of X paradox to its extreme, which is this paradox where we think that the most valuable things that humans can do are the smartest things are life. add large numbers in our heads or do any sort of white collar work. And then we totally take for granted our fine motor skill and coordination. But from an evolutionary perspective, it's the opposite.

Evolution has optimized fine motor coordination so well, and even if you look at robot hands, the ability to open a door is still really hard for robots. Meanwhile, we're seeing this total automation of coding and everything else that we've seen is clever. The really scary future is one in which AIs can do everything except for the physical robotic task.

in which case you'll have humans with like AirPods and like glasses and there'll be some robot overlord controlling the human through cameras by just like telling it what to do and like having a bounding box around the thing you're supposed to pick up. And so you have like human meat robot.

And not necessarily saying that that's what the AIs would want to do or anything like that. If you were to be like, what are the relative economic value of things? The AIs are out there doing computer programming and the most valuable thing that humans can do is be amazing robots.

Now, that being said, I think Moravec's paradox is a little bit fake. I think the main reason that robots are worse at being a robot than they are at software engineering is the internet exists for software engineering. GitHub exists. there is no equivalent thing like if you had

all mocap of everyone's actions as they were going about their daily lives for some reasonable fraction of the human population. Robotics is also... close to solve like like like on track to be solved at the same rate that software engineering is on track to be solved um so this is only like This vision is only like a sort of decade-long section. That's still a pretty terrible decade.

Imagine the world where people have lost their jobs, you haven't yet got novel biological research that means people's quality of life is dramatically better, you don't yet have material abundance because you haven't actually been able to action the physical world. in a necessary way. You can't build dramatically more because building dramatically more takes robots, basically. And people's main comparative advantage is as fantastic robots. is like a shocking, shocking world.

I mean, for the perspective of an average human, I think it actually might be better. Your wages will be higher because you're the complement to something that is enormously valuable, which is AI labor. And like, you know... a decade or two on, like, the world is fantastic, right? Like, you truly, like, you need robotics to solve and you try to get, like, you know, like, radical abundance, basically.

provided that you have all the policies set up necessary to permit building. You end up with that same change from... you know like the before and after photos of shanghai yeah yeah we're like 20 years on it's like this dramatically transformed city like a lot of places in the world probably end up like that right um over that two decade period but we need to make sure like one do our best to estimate is this actually what is on track to happen like build

sweet bench, but for all the other forms of white collar work and measure and track. That's a great thing that government should be doing, by the way, is like trying to break down the sort of functions of their economy into measurable tasks and figuring out where, what does the curve actually look like for that? Because they might be a bit shocked by the progress there. You know, there's no sweet bench for tax, like tax eval.

And, uh, then... I don't have all the answers here, but figuring out a way to share the proceeds of this economy broadly across people, or invest heavily in robotics and collecting data so that we get robotics faster, we get material abundance faster, invest in biological research so that we get all that faster. I basically try and pull forward the radical upside, because otherwise you have a pretty dark section.

I think one thing that's not appreciated enough is how much of our leverage on the future, given the fact that our labor isn't going to be worth that much. comes from our economic and political system surviving. for your million-ext S&P equity to mean something, for your contracts to mean anything, for the government to be able to tax the AI labor and give you a UBI off of that.

That requires our legal institutions, our economic institutions, our financial rail surviving in the future. The way in which that likely happens is if it's also in the AI's best interests that they follow those rails. And by AI, I don't mean some monolithic single AI. I just mean like firms which are employing AI and becoming more productive as a result.

You don't want to be in a position where it's so onerous to operate in our system that you're basically selecting for firms who either emigrate or who are doing black market stuff, etc. Which means, I think, you want to make it super, super easy to deploy AI, have the equivalent of special economic zones, etc.

Because otherwise you are just surrendering the future outside of any control that you might have on it. One of the reasons, by the way, that I worry about... turning AGI into a national security issue or having it have extremely close ties with the government, the Manhattan Project thing, is that it disproportionately redirects

the use of AI towards military tech and the mosquito drones and whatever. And also naturally puts other countries in the same frame of mind, right? If we're developing the mosquito drones, why would China not develop the mosquito drones? Um, and that just seems like a zero sum race and not to mention a potentially catastrophic one. Yes. Whereas like, You know, like, compute will be limited. You know, we'll need to disproportionately accelerate some things.

To the extent it just remains totally like a consumer free market landscape. It just seems more likely that we'll get the glorious transhumanist future where they're developing the things that make human life better. Yes, I agree. The case where you end up with two national projects facing off against each other is dramatically worse. We don't want to live in that world. It's much better if there's stays. A free mock-ups, I speak. Yeah, yeah, yeah.

Okay, I want to take issue with your claim that... Even if with the algorithms of today, if we just collect enough data... that we code automate white-collar work. First, let me get an understanding of what you mean by that. So do you mean that we would do the analogous thing of free training with all the trajectories of everything people do on their jobs?

Could you make, either manually or through some other process, some RL procedure based on the screen recordings of your white-collar worker? What kind of thing are you imagining? I mean, like a continuous distribution of this stuff. One important mental model to think about RL is I think as the task gets more

There is some respect with which longer horizon or better task, if you can do them, if you can get that reward ever, are easier to judge. So again, it's come back to can you make money on the internet? That's an incredibly easy reward signal. to judge.

But to do that, there's a whole hierarchy of complex behavior. So if you could pre-train up to the easy-to-judge reward signals, like does your website work, does it go down, do people like it? There's all these reward signals that we can respond to because we have a long... we can progress through these long enough trajectories to actually get to interesting things. If you're stuck in this regime where you need a reward signal every five tokens, it's a way more painful and long process.

pre-train on every like screen in america then probably the like rl tasks that you can design are very different to like if you could only like take the existing internet as it is today. And so, like, how much of that you get access to, like, changes the mix. Interesting. So as we're training them on longer and longer horizon tasks,

and it takes longer for them to get any signal on whether they successfully completed that task, will that slow down progress because it takes more compute per task? I do think there's this notion, the longer, the harder task. the more training is required and I'm sympathetic to that naively but We as humans are very good at practicing the hard parts of tasks and decomposing them.

And I think once models get good enough at the basic stuff, they can just rehearse or fast forward to the more difficult part. I mean, that's definitely one of the big complexities, right? As you use more compute and as you train more and more difficult tasks, I mean, I don't know, your rate of improvement of biology is going to be somewhat bound by the time it takes a cell to grow. in a way that your rate of improvement on math isn't, for example. So, yes.

but I think for many things we'll be able to paralyze widely enough. and get enough iteration loops. Will the regime of training... new models go away? Will we eventually get to like, you've got the model and then you just keep adding more skills to it with RL training? That depends on whether or not you think

there's a virtue in pre-training a new architecture. Basically, if you make some architectural change, then you probably need to do some form of retraining a new model. How does the fact that... If RL requires a bunch of inference to do the training in the first place, does that push against the thing you were talking about where we actually need a bigger model in order to have brain-like energy? But then also it's more expensive to train it in RL, so where does that balance out?

I think we've got to drink the better lesson here. There aren't infinite shortcuts. You do just have to scale. Something's going to have a bigger model and pay more inference for it. If you want AGI, then that's what you're going to pay the price of. But there's a trade-off equation here, right? There is science to do, which everyone is doing, of what is the optimal point at which to do RL. because you need something which can both learn

and discover the sparse reward itself. So you don't want a one-per-hour model. Useless, even though you can run it really fast. You also don't want a 100T model because it's super slow. Password RL. And the marginal benefit of its learning efficiency is not worth it. So there's a pretty good frontier here. What's the optimal model size of your current class of capabilities and your current set of RRL?

Yeah, and even in the last year, there's been much more of a factor of the inference cost, right? So just explicitly, like, the bigger the model, the more expensive it is to do a forward pass and generate tokens. And the calculus used to just be... should I allocate my flops to more training data or a bigger model? And now another huge factor is

How much am I actually going to do forward passes on this model once it's trained? Yeah, my total pool of compute. How do I allocate that across trained data, compute, and inference, compute for the RL training. And then even within inference, there's all this research on, well... What strategy should I use? Should I sample 10 and take the best? Do I do this sort of like branching search, et cetera, et cetera. And so with RL, where you're sampling a whole lot of tokens.

you also need to factor in the ability for the model to actually generate those tokens and then learn and get feedback. Okay, so if we're living in this world, what is your advice to... Somebody early in their career or a student in college, how should they be? What should they be planning on doing? Yeah. So I think, once again, it's worth considering the spectrum of possible worlds and preparing yourself for that.

Action that I think is like highest EV in that case is you are about to get dramatic, at a minimum, you are about to get dramatically more leverage. You already have. Like already the startups in YC are writing huge amounts of their code with Claude. So what challenges, what causes do you want to change in the world with the added leverage? Like if you had 10 engineers at your beck and call, what would you do? Or if you had a company?

at your beck and call? Like, what would that enable you to do? And what problems and domains suddenly become tractable? That's the world you want to prepare for. Now, that still requires a lot of technical depth.

Obviously, there is the case where AI just becomes dramatically better than everyone at everything, right? But for at least a while, probably, there is like... advantage i think jensen actually talked about this in an interview where he's like you know i have like a hundred thousand general intelligences around me and i'm still like somewhat useful um because i'm there like you know directing the values and like

asking them to do things. I still have value even though I have 100,000 general intelligences. And for many people, I think that will still be true for a fair while. And then as the AIs get better and better and better and so on, eventually, no. Again, prepare for the spectrum of possible worlds, because in the event where we're just totally out-competed, it doesn't matter what you do. But in all the other worlds.

It matters a lot. Get the technical depth. Study biology. Study CS. Like, really think hard about study physics. Think hard about what challenges you want to see. Yeah. That's a lot of topics. You can now. You can. It's so much easier to learn. That's right. Everyone now has the, like, infinite perfect treat-up. Yeah, yeah, yeah. It's definitely been helpful to me. I would say some combination of like...

Get rid of the sunk cost of your previous workflows or expertise in order to evaluate what AI can do for you. That's right. And another way to put this, which is fun, is just be lazy. in so much as figure out the way that the agent can do the things that are toilsome. Ultimately, you get to be lazier, but in the short run, you need to critically think about the things you're currently doing and what an AI could actually be better at doing, and then go and try it or explore it.

Because I think there's still just a lot of low-hanging fruit of people assuming and not writing the full prompt, giving a few examples, connecting the right tool. for your work to be accelerated and automated. Yeah, yeah. There's also the sunk cost of feeling like since you're not quote-unquote early to AI that you've sort of missed the boat and you can't like...

I think, I mean, I remember when GPT-3 came out. So backstory on the podcast, when I graduated college, I was planning on... doing some sort of ai rapper startup um and the podcast was just like a gateway into doing that and so i was trying out like different things and at the time i remember thinking oh 3.5 is out and people are like

I'm so behind on the startup scene here or whatever if I wanted to make my own rapper. I mean, maybe the idea of the rapper was inadvisable in the first place, but every time feels early because it's an exponentially growing process. Um,

And there were many things, many ideas, which are only becoming possible now, right? Yeah, exactly. It's that product experience I talked about before. Products are literally obsoleted. You need to constantly reinvent yourself to stay at the frontier of capabilities. But do you remember? I had a really shitty idea, and I gave you a call. I don't know what it was.

It was like, I think it was like rag for lawyers or something. Anyways, I think one of our first interactions was I'm like, hey, what do you think of this idea? And you're like, I think the podcast sounds promising. Which I appreciate. I got slightly annoyed at a friend recently who I think is really talented and clever and interested in AI, but has pursued a biology.

And I just kind of tried to shake them of like, you can work on AI if you want to. I mean, I think... humans are artificial, not artificial, are biological general intelligences. where a lot of the things of value are just very general. And whatever kind of specialization that you've done maybe just doesn't matter that much. I mean, again, it gets back to the sunk cost. But so many of the people, even my colleagues at Anthropic,

are excited about AI and they just don't let their previous career be a blocker. And because they're just like innately smart, talented, driven, whatever else. they end up being very successful and finding roles. It's not as if they were in AI forever. I mean, people have come from totally different fields. And so don't think that you need permission from some abstract entity to like...

get involved and apply and be able to contribute. If somebody wanted to be an AI researcher, like, right now if you give them an open problem or, like, the kind of open problem that is very likely to be the... I'd be quite impressed. What would it be? I think that now that RL's like come back. Papers building on Andy Jones' scaling board lengths, like scaling walls for board games are interesting, like showing that you can...

Investigating these questions like the ones you asked before, where you're like, oh, is the model actually learning to do more than its previous Parsec K, or is it just discovering that? Exploring questions like that deeply, I think, are interesting. I'd be very curious to see how much the marginal increase in meta-learning from a new task or something. I mean, on that note, I think model diffing has a bunch of opportunities.

Also, people say, oh, we're not capturing all the features. There's all this stuff left on the table. What is that stuff that's left on the table? If the model's jailbroken, is it using existing features that you've identified? Is it only using the error terms that you haven't captured? I don't know. There's a lot here. I think Matt's is great. The Anthropic Fellowship has been going really well. Goodfire, Anthropic, invested in recently. They're doing a lot of interpretability work.

or just apply directly to us. Anything to get your equity up, huh? There's just so many interpretability projects that are like, there's so much low-hanging fruit, and we need more people, and I don't think there's much time. I also want to make a plug for performance engineering. I think this is one of the... best ways to demonstrate that you have the raw ability to do it. If you made an extremely efficient transform implementation on TPU or Trinium or like in CUDA,

then I think there's a pretty high likelihood that you'll get a job. There's a relatively small pool of people that you can trust. completely own end to end the performance of a model. And if you have broad, deep... electrical engineering skills. I think you can probably come up to speed pretty fast on the accelerator stuff. Yeah, if you can come up to speed reasonably fast.

And it teaches you a lot of good intuitions of the actual intricacies of what's going on in the models, which means that you're then very well placed to think about architecture and this kind of stuff. One of my favorite people in thinking about architecture around Anthropik at the moment actually came from a heavy GPU kernel programming background, just noticing it's announced really deeply and can think about the trade-offs really well.

This was fun, guys. Thanks. Great to be back. I hope you enjoyed this episode. If you did, the most helpful thing you can do is just share it with other people who you think might enjoy it. Send it to your friends, your group chats, Twitter, wherever else. Just let the word go for it. Other than that, super helpful if you can subscribe on YouTube and leave a five-star review on Apple Podcasts and Spotify.

Check out the sponsors in the description below. If you want to sponsor a future episode, go to thawarkesh.com slash advertise. Thank you for tuning in. I'll see you on the next one.