How Agents Decide: Goodfire's Eric Bigelow on Critical Tokens, Phase Shifts, & In-Context Learning

Goodfire researcher Eric Bigelow joins the show to investigate how large language models arrive at decisions at the level of mechanistic interpretability. Drawing on his research into forking paths, Eric explains that model reasoning functions as in-context learning from sampled tokens, where outcome distributions can abruptly collapse at single, critical tokens. The conversation explores the performative nature of chain of thought in reasoning models like DeepSeek-R1, the impacts of reinforcement learning, and widespread reward hacking observed in frontier models like Kimi K3. As confidence in chain-of-thought monitoring declines, understanding these underlying decision mechanics becomes essential for evaluating model alignment. For full show notes, links, and references, read the episode page:https://www.cognitiverevolution.ai/how-agents-decide-goodfire-s-eric-bigelow-on-critical-tokens-phase-shifts-in-context-learning/ Sponsors: Parallel: Parallel provides enterprise-grade web search APIs for AI agents, offering the optimal balance of quality, speed, and cost. Get started for free at https://parallel.ai/tcr Claude: Claude is the AI collaborator for problem solvers, helping with writing, coding, financial models, strategy, and more. Get started with Claude and explore Claude Pro at https://claude.ai/tcr Tasklet: Tasklet empowers your business with AI agents that connect to your tools and automate recurring workflows with no code required. Visit https://tasklet.ai and use code cog rev for $50 in free credits CHAPTERS: (00:00) About the Episode (03:53) Forking paths in reasoning (Part 1) (15:53) Sponsors: Parallel | Claude (18:47) Forking paths in reasoning (Part 2) (19:40) Dynamics of in-context learning (Part 1) (28:16) Sponsor: Tasklet (29:53) Dynamics of in-context learning (Part 2) (29:56) Continual learning and interpretability (38:55) Shifting distributions through RL (46:41) Selecting models for research (57:11) Cognitive science and LLMs (01:07:35) Modeling beliefs in AI (01:16:22) Phase shifts and uncertainty (01:31:49) Uninterpretable chain of thought (01:45:57) Automating research with Silico (01:50:57) Safety and alignment frontiers (01:57:07) Episode Outro (02:00:02) Outro PRODUCED BY: https://aipodcast.ing SOCIAL LINKS: Website: https://www.cognitiverevolution.ai Twitter (Podcast): https://x.com/cogrev_podcast Twitter (Nathan): https://x.com/labenz LinkedIn: https://linkedin.com/in/nathanlabenz/ Youtube: https://youtube.com/@CognitiveRevolutionPodcast Apple: https://podcasts.apple.com/de/podcast/the-cognitive-revolution-ai-builders-researchers-and/id1669813431 Spotify: https://open.spotify.com/show/6yHyok3M3BjqzR0VB5MSyk

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Nathan Labenz?

  Hello and welcome back to the Cognitive Revolution. Today, my guest is Eric Bigelow, member of technical staff at unicorn mechanistic interpretability startup Goodfire. The story behind this episode is unique. After recently speaking with Bronson Shane of Apollo Research, who explained that even with access to models' internal chain of thought, it is still often extremely difficult to determine how a model will decide to act, I asked Claude to survey the literature to see what the field as a whole understands about how these critical decision tokens are chosen. One name kept coming up in that thread, and it was Eric. Eric recently completed a PhD in the Harvard Psychology Department with a dissertation

• 0:43Nathan Labenz?

  titled "Toward a Cognitive Science of Large Language Models." And while some of his academic colleagues initially questioned his pivot to focus on LLMs, the fact that even elite academic institutions are now willing to engage with AIs as a sort of mind strikes me as very notable indeed. We start today with a survey of Eric's work over the last couple of years, beginning with his 2024 paper on "Forking Paths in Neural Text Generation," in which he conducted a massive resampling experiment at every token of a model's chain of thought to identify the critical tokens where uncertainty around the final answer suddenly collapses.

• 1:24Nathan Labenz?

  As so often with LLMs, some of the findings were intuitive, while others were quite surprising and at times difficult to interpret. Interestingly, Eric says that a system's decisions are actually taken not so much by the model itself, but by the process of sampling from the model's final token distribution, which means that while the stochastic parrot paradigm should definitely be retired, stochasticity still plays an important role in model behavior. Eric frames much of what we are seeking to understand about LLMs in terms of in-context learning, emphasizing that few-shot prompts are just one clear example of how much runtime context can shape model behavior.

• 2:06Nathan Labenz?

  His bet is that this flexibility means that improved harnesses and memory systems will outcompete per-user fine-tuning, and thus that our focus of study should remain on major foundation models. Along the way, we also discuss why 7 to 8 billion parameters currently seems to be the interpretability sweet spot, why Kimi K3 was a step function for studying frontier level coding and reward hacking, and why, for now at least, interpretability outside of the frontier model companies does meaningfully depend on Chinese open weights models. We also get Eric's thoughts on why chain of thought is getting so weird, including his perspective on the chain of thought dialect and the strange language that Astro uses

• 2:48Nathan Labenz?

  to communicate with subagents, plus his pro tips for using Goodfire's Silico Research Agent, and how he hopes future AI product interfaces will begin to resurface model uncertainty to users. The bottom line from this conversation and others I've had recently on the future of AI monitoring and control is that chain of thought is increasingly unreliable, and that because these issues seem to correlate not just with RL training, but with the power of the raw pre-trained model itself, this trend might be quite difficult to reverse. That in turn means that an awful lot, including our ability to understand why agents take the actions they do, may rest on interpretability.

• 3:33Nathan Labenz?

  Eric is a pioneer of this research, but as you'll hear, tons of work remains to be done. With that, I hope you enjoy this overview of what we know and how much we have left to discover about how AIs decide what to do with Eric Bigelow from Goodfire. Eric Bigelow, member of technical staff at Goodfire, welcome to the Cognitive Revolution.

• 3:58Speaker 01

  Thank you very much, Nathan. Excited to be here.

• 4:00Nathan Labenz?

  Yeah, I'm excited for this conversation, too. I think this might be the first time in the history, and there's been like 400 episodes of the podcast at this point, that I went to AIs looking for an answer for a very specific question and came back with a name, and that name was yours. So the kind of upstream conversation I had of this one was with Bronson Shane at Apollo, who has the dubious distinction perhaps of having read maybe more chain of thought reasoning than like any other living human, at least any living human who can talk about it all publicly. And one of the big takeaways I had from that conversation was he said, you look at these

• 4:40Nathan Labenz?

  chains of thought, and the models are thrashing around, doing a sort of linearized tree search of possibility, and then at some point, they make a decision. And even having read all the chain of thought that came before that, it's often like very not obvious why they made the decision that they made, and you really couldn't have predicted it. And in a way, that's like kind of human-like, right? It's how do I make my decision? I'm not so sure I have great insight into that all the time either. But this prompted the question, okay, at some point, after all this stuff, all this reasoning, all this, again, trying to reason about what the greater wants and all that stuff, eventually we come to a passage, and it's like, and so the answer is, how does that next token get chosen?

• 5:25Nathan Labenz?

  That's what I went to the AIs to ask. Asked for a literature review, and Eric Bigelow came back as the through line through that research. And I was like, I know that company. I should get to know that guy. So what I want to do, I think you may know as much or more than any other living human about how these decisions are made, and I'm afraid that's maybe not still all that much. But what I hope to get out of this conversation is a little intellectual history of your work and like the up-to-date understanding of how the hell do these things make decisions? What do we know about that? And what is our plan to figure out how to understand it better? Because obviously, they're making more and more consequential decisions all the time. Does that sound?

• 6:08Speaker 01

  I'm very honored to be the first guest chosen by the AIs. I always, I feel like they, when I talk to Claude, it seems to really get excited about my research, but I'm sure everybody feels that way. I think, I think actually talking, having some conversations around this podcast and some of the things you're asking about makes me think that I actually should devote more effort to this exact question of how decisions are made. But I'll give a quick overview of a project that I think is really one of the most informative projects in how I think about how LLMs work at a basic level, and that's my project Forking Paths in Neural Text Generation.

• 6:52Speaker 01

  And for that, basically, so this was two years ago with, at the time, it was one of the GPT 3.5 models. And what we do is we take reasoning chains, we give it a chain of, we give it a question and a chain of thought prompt and have it reason through, you know, step by step. And then every single token in its reasoning, we resample all these completions. So we resample like 30 different rollouts at every single token. And then from every one of those rollouts, we see what final answer it landed on. And you can aggregate these into various distributions and then do analysis on these distributions. But I think the most striking part of this is if you arrange it like a multivariate

• 7:34Speaker 01

  time series, you see these really interesting dynamics that are happening. And I kind of take inspiration from this, from some of the learning dynamics work that was coming out around that time, like around Grokking and things like this, where there are like these really sharp phase transitions in training, and the model just seems to suddenly learn something. And it turns out that you actually have these during reasoning and in-context learning as well, where over the course of reasoning, there will be certain points in reasoning that the model was initially sort of, it had 50% chance of this answer and 30% chance of this answer. And then there's some point that it hits where suddenly the distribution changes really dramatically, and often the distribution just kind of collapses onto a final answer

• 8:19Speaker 01

  as the model sort of, you might say, decides what it's going to say next. Although I think, I use the word decides here with maybe scare quotes around it because I think there's some nuance to what that means. But one of my big takeaways from this, and I think this also just kind of falls out of thinking about the architecture, but it's nice to have results that really support this, too, is that models, even to some extent when they plan out their responses, what they're actually going to say at every single moment is really determined by like the flip of a coin, which is what token will it generate next? And there is sort of an interface that was really standard and with OpenAI at least at the time, which I think is unfortunately no longer present, and I haven't really seen

• 9:04Speaker 01

  standard interfaces, which is where you could see the log probabilities of all the tokens in a text output you got, and you could kind of hover over different tokens and see what are the alternate tokens that could have been generated instead. And if you do this with a lot of text generation, you might realize that you get this one text answer when you ask some question, but really, there actually could have been a lot of different answers. And you might ask, what would have happened if this word was different from this word? And what we found is that there are certain key words where sometimes the words, if you were to sort of like look at it like a judge would, you might guess this, like the first time that a final answer is mentioned in the reasoning chain. But then there's other times where this distribution changes really dramatically on

• 9:46Speaker 01

  some sort of arbitrary word, even like an open parenthesis for like, you know, there's one example we have in the paper where it's like kilowatt hours, open in parentheses, KWH. And, you know, you never expect that to really change how the final answer plays out. But if you actually go to that exact token with this model and you just resample a bunch of text generations, you see that in cases where it's not an open parenthesis, where it starts to be some other just arbitrary word, it ends up being a different final answer. And so I think--

• 10:20Speaker 01

  This thinking about decisions in LLMs or what they're going to say next, very much being determined by sort of chance and by whatever tokens are sampled, is really, I think, the foundation for how I think about a lot of things in LLMs. And I think it sort of shapes how I think about beliefs and models and sort of what's considered like a factor hallucination can be dependent on what tokens were generated before it. And I think one of the properties of reasoning that really comes into play here is self-consistency. So if there's a long reasoning chain that is self-consistent and it generates some token earlier in the chain of suppose that this was the year instead of this was the

• 11:01Speaker 01

  year, if it thinks that the year is 2024 instead of 2026, actually, the answer to a question might be very different. And there's cases like this where it's like you might think, well, obviously, there's like a ground truth. If I'm asking, who is the current head of state? There is a current year, which right now is 2026, and so the model was hallucinating. But I think there's also a lot of use cases with LLMs where we don't really know what the ground truths are. And there's sort of this branching space of possibilities that we actually want to explore naturally when we're interacting with a model. And like, for example, if I'm working on a scientific research project and I want to sort of talk to my LLM about like, what should I do next?

• 11:45Speaker 01

  Or what did my experiment results mean? I might really want to kind of like iterate and understand what's this full space, this whole tree of possibilities that it could have generated? And what would have been different if it had assumed A instead of B? And so to go back to your question, I think the simple answer I have is that I think this decision really happens during sampling is my biggest part of how I think about this is that, again, every token that's generated by the model. To... One of the other big influences in my work is thinking about in-context learning and how models adapt to context. And the way I think about reasoning is that models are basically in-context learning

• 12:28Speaker 01

  from things that they have generated already. So the words that they generate, there's sort of this stochasticity that's happening. There's these coin flips that are happening as it's, you could say deciding, but it's really just a sampling from a distribution of these alternate tokens. But depending on what token is sampled, it in-context learns and there's this context dependence where then it'll generate a set of reasoning that is consistent with that. And I think that like, I'm pretty bullish around like, Mecanturp shedding more light, I think, on really answering this question of how decisions are made. But then, I think, I think...

• 13:12Speaker 01

  So I said that this work was done a couple of years ago, and I think that modern reasoning models are a little bit different than chain of thought reasoning was then. And some recent projects I've worked on, even with like Mecanturp of reasoning models, have kind of changed a little bit how I think about what these models are doing. And so I think there's still context dependence, but actually, as you mentioned, reasoning models, I think, often tend to do this like really like linearized search where they enumerate all these different possibilities in context. And then at the end, they sort of just go back and pick one of the things that they said before. So I think this might change exactly how this happens.

• 13:54Speaker 01

  From what I've seen, the uncertainty dynamics, if you do this like forking path analysis on reasoning models, is often a bit smoother, but there are still sometimes these forking points where if one token is generated different from another, or if the model was really doesn't, you know, it enumerates all these reasoning chains and none of them really seem that much more promising than another, then eventually it does sort of, there are cases where it will just say, okay, the answer therefore is B, C, and it actually could have been different depending on what was generated. But I think there could still be more work done in understanding the internal processes that lead to that, that point of choice even.

• 14:35Speaker 01

  And another project that I'm on has shown that there are, at least with some other toy domains, there are these cases of like neurons that seem to like track model confidence, where you can even like steer and make the model more or less confident generically about whatever concept it's learning. But yeah, I think we really need more interpretability work in this vein. And I think in-context learning and reasoning are unfortunately vastly neglected in the world of interpretability compared to how, I think, how defining they are for how we think about lots of the high-level things that LLMs do.

• 15:15Nathan Labenz?

  Yeah, I mean, when you talk to the guy who's read more chain of thought than anyone and you keep in mind that a huge amount of the frontier companies' plans right now rests on chain of thought monitoring as like the way we're gonna keep track of what the AIs are doing and if they're doing the right or wrong thing. And then you hear that, yeah, even having read it all and being like fluent in these like strange internal dialects that they use, I still can't tell you why it chose what it chose. It's, oh, yikes. Where's the red phone to call the interpretability department? Hey, we'll continue our interview in a moment after a word from our sponsors. Today's episode is sponsored by Parallel, where agents find answers.

• 15:58Nathan Labenz?

  Most engineers today closely follow new model releases, but don't pay nearly as much attention to their agent's most important tool, web search. If you're like me, your agents are often conducting hundreds of searches per day. And while the unit cost is small, over time, this does start to add up. Before starting to use Parallel, I calculated that my agent's search bill would total roughly $500 this year. The good news is that I just had my agents conduct a systematic test, and I found that Parallel's fast mode, which costs just $1 per 1,000 queries, worked just as well as my previous default provider. And switching to it will save me roughly 80% of my search bill going forward.

• 16:39Nathan Labenz?

  Parallel's infrastructure is enterprise-grade, and their suite of APIs offers a range of Pareto optimal options that allow you to choose the right balance of quality, cost, and speed for your needs. So whether you're building voice agents that need 200 millisecond latency or long-horizon agents that need deep research, adding Parallel to your agent's toolkit is a no-brainer. Get started for free at parallel.ai/tcr. That's parallel.ai/tcr. Today's episode is brought to you by Anthropic. By now, you know my story. Claude drafts my intro essays, and I rewrite them.

• 17:21Nathan Labenz?

  Not because the drafts are bad, but so I can stand behind everything I publish. Well, I have an important update. Claude Fable 5 is the first model to have me rethinking my rule. Today, I now think co-authorship, not sole ownership, should often be the goal. Where the model excels, rewriting its work can be more about vanity or a misplaced sense of duty than integrity. I feel it most in songwriting. I'm no lyricist, but I'm good with a song concept, and Fable writes some amazing verses. I give it feedback on its misses, and I push it to aim for higher inspiration, add layers of meaning, optimize syllable density, and above all, write a hit song.

• 18:06Nathan Labenz?

  These days, I get compliments on just about every song we write together. Claude is the AI for problem solvers. It's the collaborator that understands your entire workflow and thinks with you, not for you. Whether you're debugging code at midnight, building a financial model, or strategizing your next business move, Claude extends your thinking to tackle the problems that matter. For problems worth solving, get started with Claude at claude.ai/tcr. That's claude.ai/tcr. And check out Claude Pro, which includes access to all of the features mentioned in today's episode. Once more, that's claude.ai/tcr.

• 18:47Nathan Labenz?

  I think there's a few different directions you started to touch on there that I would like to follow up on. Maybe for starters, let's go on this relationship between in-context learning and other things, right? Like, there's been a lot of work that's looked at, if not, this is maybe too strong, but sort of an isomorphic relationship between in-context learning and fine-tuning. I think that's like especially true at Laura scale fine-tuning, where you can localize the weight changes. Maybe it's also true, even if you're doing like full-weight fine-tuning, I'm less sure about that.

• 19:32Nathan Labenz?

  How should we understand that relationship? What do you think is the most important that may not be obvious?

• 19:40Speaker 01

  So I think the most important thing I'd say about in-context learning that may not be obvious for people, especially if you're just getting started learning about it, is that I think when people say in-context learning, historically, it's meant this sort of traditional few-shot, give a few shots of input-output examples and see how the model behavior changes. And you can do these like scaling curves, for example, where you like increase the number of shots that you give and see how model performance improves as a function of this. But then I think...

• 20:14Speaker 01

  The way I think of in-context learning, and I think there's a few other people that have echoed this, and one paper that I like around this is by Andrew Lamban and others on the, I think it's called the broader spectrum of in-context learning, where I think, what I think of in-context learning as capturing is everything the model is doing to adapt its behavior that's not happening with in weights learning. And so this is how it's changing its representations as it reads each new word of input text. And this also applies to reasoning, where with reasoning, it's generating text and then using that to determine its following behavior.

• 20:54Speaker 01

  And so I think, I think this view of in-context learning is sort of being really ubiquitous. I think it gives some teeth to thinking about prompting, where like, there's this term prompt engineering that is sort of a little bit of like a dirty word, I think, in the scientific community, where it feels like people just come up with like a bunch of hacky solutions that make the models go better, you know, tell it you're a brilliant software engineer and it works, and it writes code better, these sort of things, although these are maybe a bit less prevalent today, because maybe we've kind of, I don't know, fine tuned some of that, some of that distribution shift just into the models. But I think, I think there's a way that it really defies comparison to like this kind of in weights learning,

• 21:37Speaker 01

  where really prompting is, that's not going to disappear no matter how much fine tuning you give. You're going to have different contexts that you give the model, and you want it to adapt its behavior dynamically, right? Like, I think, even if you, I think, fine tune a model enormously, like you still want it when it's like reading, when it's reading Harry Potter for the first time, I mean, of course, it's in the training data, but suppose there is another, you know, story it was reading for the first time, you'd want it to be updating its beliefs internally about what's going on in the story, who are the good guys and the bad guys, who is, who is happy and who is sad at different times. And really, like, I have another paper around this sort of belief updating and representation change and stories, and I kind of like this metaphor, because I think it, it's easy

• 22:22Speaker 01

  to understand that, you know, as you read a story, your beliefs should change. There's a bunch of work around analyzing the sentiments of stories, and one of the inspirations from that project was this famous talk by Kurt Vonnegut on the shape of stories, where he's describing like these different stories, like Cinderella starts, the Cinderella's very sad, everything's horrible, and then she gets invited to the ball and everything's great for a little while, and then the, and then the magic sort of falls apart and she has to go back to her normal life, but a bit better than she was before. And then finally, at the end, when, when the prince finds her, you know, she has an infinitely happy ending, or however he says it. And I think, like, stories have shapes, and really

• 23:06Speaker 01

  all contexts that we give models also have shapes that are dynamic and temporal. And I think this is something you just, it's, it's hard to really compare it to fine tuning or to these other ways of changing representations. And I think, more broadly, I think that this is a piece of the puzzle, the thinking about interpretability of LLMs and not just feed forward networks, is that everything that LLMs are doing is happening dynamically. They're updating their beliefs, they're shifting representations internally, they can learn new representations in context. And this doesn't just happen when you give a few input-output examples. This is happening all the time. This is happening over a conversation, this

• 23:49Speaker 01

  is happening within sentences, this is happening during the period where a model is reasoning or generating a really long input or output text, it's sort of changing its representations dynamically to adapt what it's doing in the moment. And, yeah, I think, I think a lot of work kind of misses these dynamic changes that are happening. I think a lot of, a lot of interpretability work sort of focuses on these like templates, where there's just sort of a couple, a couple of variables that are changed in the input, and behavior is mentioned on like one token or one, or one, one sort of output variable. And really

• 24:30Speaker 01

  what LLMs are doing is, I think, much more dynamic and adaptive all the time. One other point I want to, I want to sort of raise on this view of in-context learning is that, something I was inspired by too, at the time with, with Grokking and this kind of work, was this idea that, that if you look at learning curves over a huge data set, they'll be very, very smooth. And as you just keep training the model and scaling parameters, things will just very, very smoothly increase as you, as you train it for longer and longer, or errors will decrease, rather. But then, if you look at specific problems, specific problems can actually change very sharply. There actually can be

• 25:10Speaker 01

  these phase transitions. And, and I think it was, it was Neel Nanda who had this quote of like, phase transitions are everywhere, if you just looked closely enough. If you broke your data set down into like all the, into just individual tasks, and you looked at individual tasks, there'd be a lot of stepwise changes where the model really seemed to just suddenly learn something. And I think in-context learning is like that, too, where if you look at aggregate, if you look across a large data set, things might look smooth, maybe deceptively so, but actually individual cases, like an individual story, have these really nonlinear dynamics. And curves can change very sharply, things can

• 25:50Speaker 01

  change after even just one, one token or one sentence. And other properties can change, not just, you know, suddenly learning something, but beliefs can change, beliefs can go up and down and follow all sorts of, all sorts of weird twists and turns. And so I think, another, another thing that I think is sometimes missing from views of in-context learning is this, you know, zooming onto one example sort of problem. And something that I think is also like, I'm, I'm really curious if the, if the person you had earlier, who's looked at all these chains of thought, if you looked at, like, resampled outputs for

• 26:33Speaker 01

  the same model, or kind of done some of this, like resampling within a single chain of thought. Because I think sometimes you can get a lot from looking at patterns by just like looking at like a ton of different like rollouts and just like trying to understand like how the model thinks in some general way. But then I think, like, for an individual prompt, there's sort of this like, really elaborate landscape, just in how the model will think about one problem. And I think sometimes you can only get this by looking at a lot of different resampled outputs for that same problem and seeing like, where are the patterns, what kinds of things are similar or different. Like, if you, if you give models, depending on which model you, you have, if you give them a very open-ended

• 27:17Speaker 01

  prompt, like, generate a story, sometimes you'll notice these really funny patterns, where they'll write very different story, like, things will be superficially very different, where it'll kind of start out like very different, but then they kind of, like, I asserted out that Forking Paths project by just, tell me a story about a boy named John, and like, half the stories involve John having a dog, and they go on an adventure, and a whole lot of those involve them eventually finding pond or a waterfall or something. And then like, sometimes the dog saves the day, or sometimes John saves the day. But I think these are all these different patterns you see for outputs for these same input example. And so I think, I think we need to do more of that, like, looking at, like, a lot of

• 28:02Speaker 01

  resampled outputs for the same input examples to kind of understand, like, how bumpy or smooth the landscape of in-context learning is for these models.

• 28:10Nathan Labenz?

  Hey, we'll continue our interview in a moment after a word from our sponsors.

• 28:16Nathan Labenz?

  The Cognitive Revolution is brought to you and powered by Tasklet. I've been running Tasklet agents for almost a year, and they've become a major part of how the show gets made. Before every recording, one agent researches the guest and sends me a suggested outline of questions, complete with links to all sources. Meanwhile, another agent scores the overwhelming number of pitches that I'm getting from PR firms. And that's really the core idea behind Tasklet. Run your business on agents by giving them real jobs to do on an ongoing basis. Setup is incredibly easy. Just describe the work in plain English. Research a new sales lead, update the CRM, and draft a personalized reply. Tasklet connects

• 29:00Nathan Labenz?

  to your business tools and can run around the clock, on a schedule or when something happens, like a new email arrives. Each agent works with your tools and business context, and remembers useful information across threads. You can also share agents, connections, and reusable skills with your team, so the workflows that one person figures out become something the whole company can use. No developer needed. You decide what agents can do on their own and what needs your approval, and there are no per-seat fees. You pay for usage, not for adding teammates. Start with one recurring job, like researching the people you're meeting, and then just keep building from there. Visit tasklet.ai and use code

• 29:45Nathan Labenz?

  COGREV for $50 in free credits. That's code COGREV at tasklet.ai.

• 29:54Speaker 01

  One big thing pushing toward is continual learning. And I wonder what your experience...

• 30:05Nathan Labenz?

  Trying to understand in context, learning suggests or implies about how much additional challenge full continual learning might present. I could imagine, like, I think the naive read would be like, this is going to be really hard now because we've got everybody's individual model becomes personalized over time and who's to say how different they are. But then the, maybe the counterargument would be like, we already have that within context learning because there's a lot of context and we basically already deal with that same problem. Does your experience make you feel like we're ready for continual learning in terms of our interpretability

• 30:46Nathan Labenz?

  techniques or that it will still pose like a step change increase in difficulty?

• 30:52Speaker 01

  I think, um, I think for your question of whether our current interpretability techniques will face challenges with continual learning, I think it really depends on how continual learning plays out. And it's something that I haven't like personally done research on or thought too much about in depth on like, do we need different architectures for this? And I know there's sort of a lot of people that have been debating this for a while. And as you're saying, like having personalized models or having models that are like fine tuned to your data. I think if that's the case, then there's a possibility that we'll have to have interpretability, our interpretability techniques will have to evolve as well and we'll have to have some sort of like, like, I think, I think how, how well, you know, say,

• 31:37Speaker 01

  say that you have a tool that's trained on a base model and then that model is fine tuned for 10 different people or 10 million different people. I think how, the real key question there and how well your tools will directly translate is how much representation change there is. And like, um, I have some work on this and other folks at Goodfire have some really cool work on there being these low dimensional manifolds that describe the sort of underlying representation space that models are going through as you're, as you're interacting with them. And as, and a lot of the representations are sort of defined by this low, low dimensional conceptual structure. And I think my intuition would be that as long as a lot of that structure

• 32:20Speaker 01

  remains the same, then a lot of the techniques will be pretty easy to translate. And maybe you need to have a bit of fine tuning of your probes so that they work on each person's particular fine tuned model in that world. But I think the only real challenge there is if the continual learning leads to such catastrophic forgetting that the internal representation structure of the model radically changes. But I kind of, I kind of doubt that'll happen because that, I think that will probably mean that the models will break. Like, it may be that they, there is massive internal representational reorganization, but things are still structured and the model can still do lots of like, all the old things it can do, but in new ways. But it seems to me more likely that if, if we're in a world where

• 33:04Speaker 01

  really there is this collapse of common internal structure, then the models are probably, probably breaking in that case, unless in this world of continual learning, we're using an ungodly amount of compute for every one of these 10 million people to, to train them on, you know, do, do a lot of meta learning so that they maintain their performance on all these tasks and learn entirely new representational structures. So, yeah, I think, I think in that world, it's kind of like having like a bunch of slight tweaks on a main model and we probably will have like, tools that we can directly translate to these fine tuned models and maybe just do like a little bit of fine tuning on, on, on the tools that we have and the probes and such. To the question about thinking about continual learning and

• 33:49Speaker 01

  in context, learning, my personal view, I mean, again, I don't have strong perspectives around their, around the architecture argument, but I think there's sort of a bit of a bitter lesson around like prompting and in context, learning and context engineering and harnesses and all that, that they might be all you need for quite a lot of stuff. And I have a bias towards this because this is what I work on, but I kind of, for me, the most likely outcome in the, in any short term is that we don't have custom models that are really trained on very different data, but they're actually, what's happening is it's different context, right? As you're saying, this is sort of what already happens. We have different LLMs that have different memories. And the memory is just the way I see it, a

• 34:34Speaker 01

  bunch of data that is tokens that are in context learned. Every time the model generates an answer, it, it sees this, it sees this long memory. And, um, I think that it's just, it's incredibly easy to get a lot of, a lot of juice out of that without squeezing too hard. And then you end up with these memories that are pretty interpretable. And, uh, yeah, I think, I think some variables would have to change in the equation for me to see it being really that promising to fine tune models separately on individual people's data to have customization really pay off in that way. And so maybe my, my, my personal best guess is that we're going to see more context engineering and more of this, this kind of thing, more

• 35:15Speaker 01

  harnesses, more memories. And I think that's really what we've seen in the last years with, with the rise of agents is, is a lot of, a lot of that engineering is just going into like putting a lot of stuff in context.

• 35:28Nathan Labenz?

  Yeah, it's been striking that Anthropic has never had a fine tuning product, at least that they've offered to the retail developer customer. And OpenAI has killed theirs. And what little test cases we do have around interpretability applied to different architectures have been for me like quite pleasant surprises in terms of how well techniques have translated. I'm thinking, for example, of Othello GPT and then like Othello Mamba, both were basically interpretable and seemed to have like similar representations internally. And those were recovered through like fairly similar techniques. And yeah, a long way those favorable trends

• 36:10Nathan Labenz?

  continue. On the question of kind of change in models and like, how much the beasts themselves have morphed in front of our eyes over the last couple of years, I'm old enough to remember when temperature was a feature that we got to play with at the API level. And at that time, I knew how my tokens were being sampled from the distribution, subject to like terminacy. For, there was a time, I'm also old enough to remember a time when I didn't know about GPU and terminacy. And I was like, why would I turn my temperature to zero? Do I not get the same thing every time? And that turned out to be the answer for that. But I guess,

• 36:55Nathan Labenz?

  how, how would you describe, because it sure, the temperature has gone away. It sure seems like the models are, you know, mode collapse is maybe a little bit strong, but like, they're quite overfit in some ways. Your story example is a good example of that. Like GPT-3, with its wild untamed nature, was probably in some ways a better bedtime story generator than GPT-6, which like always gives you the same kid and dog and waterfall. Even seeing that right now, weirdly across providers where, I won't bore you with it, regular listeners have heard me talk many times about how I use models to draft the intro essays for

• 37:37Nathan Labenz?

  the podcast. But in testing Astro versus Fable recently, I've also noticed like very like uncannily similar decisions in terms of how they choose to try to present an episode. And I'm like, this is not random that they're coming up with these very similar ideas. There's some like, pretty tight fitting going on. And then we also have seen like the introduction or at least the emergence, maybe is a better term of these meta cognitive behaviors where you see, oh, I think I'm off on the wrong track. Let me go back and approach this another way. The aha moments kind of thing. So, how would you describe, how would you like tell

• 38:19Nathan Labenz?

  the story of two years ago, when you were doing this like super high volume sampling and you were seeing these like pivotal tokens, how have the underlying models changed and what should we be aware of as users and as people trying to understand how decisions get made and as interpretability researchers based on just how dialed in they seem to be? And I'm also just interested, do you even know like how sampling is being done these days? It's a very in the weeds question, but yeah, tell me the story of the last two years. How do you understand how models have changed during that time?

• 38:55Speaker 01

  I'll say first that I wish I knew what the, what the sampling parameters were for the Claude API and for the latest GPT models. I think it's, it's very unfortunate that they've tucked these away. I expect that it's probably similar things happening. It's just that they want to limit how much people can reverse engineer the models and play with temperature to do a lot of systematic, like, I think a lot of the people that are doing really systematic prompting are using like a lot of, you know, hitting the Claude API with millions and millions and billions even of token requests. There is a high risk that they're doing something, you know, a little bit nefarious or trying to distill models

• 39:36Speaker 01

  or, or extract model weights in some way or another. So I think that's probably the main reason why they did this rather than that they're actually doing a different, a different sampling scheme under the surface. Although if I, I would love to learn more and I'd be happy to be surprised. As for the story for the last couple of years, so to your point that

• 39:58Speaker 01

  the earlier models were more like unleashed, as you put it. And I think with the word that comes to mind with this is something people have studied around output diversity. Where what diversity means is that if you, it's basically the same. If you give the model the same prompt and you have it generate a bunch of different outputs, how many different outputs will it generate? And I think this is something that if you, if you, there's been a lot of work showing that this disappears to some extent with supervised fine tuning. And the more, the more you post train models, the less diverse they, their outputs become. Which really isn't surprising at some level, because the earlier like completion models that weren't even chat, they were just raw completions. Like when you talk to it, it wouldn't even know that it was in a conversation.

• 40:41Speaker 01

  It's just doing text completion, next word text completion on the whole internet. It felt very much just like this raw thing that you're interacting with this like big amorphous blob of internet data. And if you just start giving it random tokens, it'll just complete those in some way that's coherent. Whereas afterwards, I think we evolved that we first went into a world of supervised fine tuning and RLHF and other post-training for chat and basically model instruction tuning, right? Is really fine tuning models to this distribution of you're going to be talking to a user who's going to give instructions and you should respond to those sort of like a chatbot should. And then RLHF I see as further like sharpening that distribution around chat.

• 41:25Speaker 01

  You should act like a chatbot and talk to people in a coherent way. That's, that's consistent with this like chat interface that everybody is using now. And then I think RL does something very, pretty, pretty different. And I think RL, I think, I think a couple of projects I've worked on have shaped a lot how I've thought about RL. As well as some of these things recently about actually looking at the reasoning of models and seeing how uninterpretable they are. And I think RL kind of, I don't know how to think about it in terms of it. Like I think it, it sort of sharpens the distribution in some ways. Or like there's some really RL'd smaller open source models that are like kind of cooked.

• 42:06Speaker 01

  Like they'll kind of like, you know, I've, I've, I've done some, a bunch of experiments recently with this Qwen 3.6, 35B, A3B like model. And it was RL'd pretty heavily. And when you give it the, when you give it this kind of prompt of like, give me a story. Two thirds of the story are about a clockmaker and a bunch of the other ones. I forget what it is, but there's like another pattern that just comes so much in the stories that it generates. But I think like, I think I've actually tried this recently with like Astro, for example. And it actually generates like a bunch of different stories, which is, which is interesting. Like it'll actually generate fairly different stories with fairly different arcs and characters.

• 42:50Speaker 01

  I mean, I haven't done like a really massive scale of sampling like hundreds or thousands of different outputs and looking them, looking at them. I, I probably ought to. But the impression I have of reasoning is that reasoning is, it kind of shifts the distribution of what language models are doing in a totally different direction. That's very different from like the human text that they were trained on. Where they, they'll, going back to this, this phrase you said earlier, that I really, I think very consistent with my thinking that they do like a linearized tree search, that they'll enumerate like all these different possibilities of how the problem could be solved. And sometimes the possibilities that they enumerate are related to each other.

• 43:34Speaker 01

  But when, there's this idea of wait, wait, being like a really like key point where the model changes its mind. But if you look at like deep seek model out, like R1's outputs, they'll say wait 50 times or more in a single reasoning chain. That's a lot of epiphanies to have when you're working through one problem. And I think this was really like, this was a solution to this problem that models were hitting at the time, which was how can they backtrack? Which is related to this forking I mentioned earlier, whereas it's, as it's generating a reasoning chain, it can randomly sample some word. And then it can't really step back from that path. It's just pushed itself down that path forever. But I think what reasoning models do is they get around that backtracking by just enumerating like every possible thing they can think of.

• 44:18Speaker 01

  And then later going back and choosing from that. And I think my impression of at least like probably the most state of the art proprietary models like GPT-6, Astro, why can it generate different stories is probably that it just generated a bunch of different possibilities in its reasoning chain. And then it's going back and looking at them. And that's where the sampling is really happening. I think this, I think maybe some of the diversity is happening within the reasoning chain itself is my way of thinking about it. But then, I don't know, I should probably do more systematic studies with RL models to, to understand this, this problem of diversity better. But I just, I think, yeah, some recent projects have shaped where I really don't. I think reasoning is almost like a misnomer for what reasoning models are doing.

• 45:03Speaker 01

  I think traditional chain of thought reasoning really was like human reasoning in some way where it's like, let's break it down step by step. And here's a set of propositions, A, therefore B, therefore C, therefore D. But reasoning model will just enumerate A, B, C, D, A, E, F, G, FGA, all these different. It's like a, it's almost like a soup of tokens they create. And then they just go back and skim out something out of that soup. And yeah, I think, I think that's my high level summary of a lot of the research that's happened in the last couple of years. And I personally, I think that in terms of raw capabilities, I think there's been like maybe some gradual change with this and a bit of extra scaling in the last few years.

• 45:45Speaker 01

  But then I think a lot of the raw capabilities were probably in these pre-chain models the whole time. And a lot of what we've done since then is finding ways to just really evoke these and get these out. And to sharpen the distribution of what models are doing. Coding agents, for example, is something where I think that just had to train models a lot doing coding to get them to behave in the right way. Even if they had learned the right things from the entire internet, internet of text. It's like this extra version of this, this additional case of this base models being just next word completion over the whole internet. And I think a lot of the agent fine tuning and getting success with agents is really polishing what distribution they're fit to.

• 46:27Speaker 01

  You're a coding agent, you're doing lots of changes, you're working with Git and Diffs all the time and you're running tests. And I think this is more than anything like a, a small distribution shift or a sharpening of a distribution.

• 46:41Nathan Labenz?

  When you do research, how do you choose what model to work with? Obviously, a lot of people are choosing based on what the compute resources they have available support. And I'm sure that's a factor for you too. But at Goodfire, you guys have significant resources and you've built some infrastructure to do interpretability on whatever Kimi K3, right? Is it a frontier scale? Is it near frontier scale? It's big anyway. How, I guess, what is your sort of mental model for what kinds of models to choose when? And when you do something on a frontier Chinese model, in your mind, is that essentially synonymous with doing it

• 47:26Nathan Labenz?

  on a American proprietary model? How much do you think of those as standing in for each other versus how idiosyncratic do you think they might be?

• 47:36Speaker 01

  So I think, I think of them as in large part standing in for each other. There are idiosyncrasies of like, Qwen really likes to think in Chinese. And if you do the low Jit lens on Qwen, sometimes there will be a lot of Chinese characters sort of in the, in the middle before it starts outputting English tokens, if you're talking to it in English. For your question of choosing the model that I work on, it is quite a privilege to have like these like big compute resources at Goodfire. And something that I'm, I think, still, still getting used to and finding ways to really like take advantage of in the best way possible. But there still is a problem of if we're doing like a lot of experiments, you know, I, I think it still would be hard to justify me

• 48:19Speaker 01

  taking up hundreds of GPUs for a substantial amount of time to do like really, really elaborate experiments on Kimi K3. If they weren't really like in line with some of the company's goals or sort of taking advantage of some time where there, there's a bit of downtime and they're underutilized. So I think there still is, I, I still do think about compute and kind of having some trade off of, you know, what's the right model scale? Where I can study the behaviors that are interesting enough. So I think generally how I choose is I pick a model like I, I think there's, if you go down too small, then you're working with something that, I don't know, is, is.

• 48:56Speaker 01

  Qualitatively different, I think, if it's like, you know, in the below a billion parameters. Depending on the task, sometimes there are still interesting behaviors you can get out. But then I think there's a lot you can learn by like training models from scratch. Where you train like 20 neural networks or transformers on a version of your task and you look at learning dynamics or you do interpretability on them. But a standard that I've always set for myself in my research is to try to study something that's kind of like the big models where you do see these really elaborate, human-like behaviors and you can just study them zero-shot in the model without having to train the model to do that. And without having to use like a facsimile of the model. And so with choosing a model, I usually, I tend towards like

• 49:40Speaker 01

  the scale of seven to eight billion or more being a sweet spot where like you can start to do good interpretability research. And I think also like the few billion models can be good if your task is like relatively simple, but I think sort of around that scale of like eight billion or more is when you can start to do this sort of like zero-shot prompting and get interesting behavior like my my story belief updating paper. Initially, we just used Llama 8 billion. And I wasn't totally sure at first how it would do, but it actually does, you know, update beliefs very dynamically as it's reading stories across like a number of different emotions and other and other attributes that we tested. And so I've been, I've been pretty impressed with models abilities at that scale. And then

• 50:25Speaker 01

  I mentioned that recently studying this Qwen 35B A3B model. And I was also, I was very, very when I, when I started before working on the model too much, I wanted to just chat with it to get a sense of what it's like. And I wasn't sure if it would be kind of like, yeah, kind of, kind of cooked, as I said, like some of the initial experiences with having it generate stories and seeing like a lot of the same stories. But then something that I just like doing with LLMs in general is talking to them about consciousness and see what see what they say and how the conversation goes. And I think that that gives me some signal.

• 51:01Speaker 01

  Maybe it's like a strong bias here, but of how well a model can, you know, think about a pretty nuanced and complicated topic. And I don't know, I was very impressed with how well this model was able to talk about the problem of consciousness and LLMs and sort of think actively about that. I mean, there's also this well documented or it's a well documented, but also a bit speculated around what the post-training data of these models is. And if they're really like, to some extent, distilling some of the, some of the proprietary models and to whatever extent that is true, that could explain why there's a lot of similarity with these models.

• 51:45Speaker 01

  But I think, yeah, that in terms of their core capabilities, they're all really trained on the same data, which is the entire internet. And the main difference, I think, between all of these models is the post-training and what is that extra special sauce that we're giving the models. And I think in terms of the moat between closed source, I personally think that's probably been the biggest moat and will continue to be is, is the post-training that they're doing. And at various levels. And I think that that is one of the things that really differentiates the proprietary models from the open source models. And, uh, and I think what that post-training does is it will make the models way better at like coding or specific things that you do care about.

• 52:24Speaker 01

  It when you're actually using the models, but I think a lot of the core capabilities will be very similar. And Kimi K3 is also this big, I think this big step function and having in terms of having a model that's really good at coding that we can study. And although I haven't studied this personally, I've, I've seen some of the other research that people at Goodfire have done recently, like with reward hacking and it's, it's I mean, it's very impressive how good the model is at coding. And also it exhibits this, you know, reward hacking, like crazy behavior that you just don't see at smaller scales. And I think does give us some signal into studying phenomena that we think are happening inside proprietary frontier models. So I think like there

• 53:07Speaker 01

  there's definitely a gap in performance between the proprietary models, the sort of, you know, OpenAI and Anthropic models. And almost everybody else with a few exceptions. But I think like in terms of the kind of work we do with interpretability, this doesn't, I think it's not too much of a blocker. The, the main blocker is when there are certain behaviors that you just really can't study, except with enough scale or post-training. Like I think like this reward hacking with with Kimi is, is like a really interesting example where it seems like just having the model pass that step function, just in terms of some, some particular behavior coding ability was very important to start also getting to see these really

• 53:50Speaker 01

  these dangerous behaviors that we want to study and understand better. That like, it's, it's really hard to get it out of a smaller model. Or if you get it out of a smaller model, it just, it may be that it's happening through a very different means.

• 54:06Nathan Labenz?

  Does that mean that interpretability research today outside of proprietary model companies is becoming dependent on Chinese models? Is there an American model that you could use as a substitute for Kimi K3 that you feel like wouldn't lose that much of the value or are we like, we have to use Chinese models if we want to do frontier interpretability in today's world?

• 54:34Speaker 01

  I think for now we do. And I think in the biggest models, I think there is this, there is a big gap, particularly in programming ability between Kimi K3 and everything else. But I think this isn't a gap that couldn't be closed easily if there was more investment in open source models from, from American companies. And I think when you go to like smaller scales, the gap probably gets a bit smaller. Like I think I, I do study the Qwen models pretty heavily and I think they are like really like state of the art for their model scales. And I'm, and I'm very glad that they exist and I can study them.

• 55:14Speaker 01

  But I also think that if it, if I was stopped from using Qwen on the 35 billion scale and had to study the Gemma models instead, it probably wouldn't slow me down too much, maybe a bit. I think right now the gap really exists at the, at the top of the food chain. But then like, yeah, I think if there ever was like a ban on using Chinese models, I would really hope that some American companies would step up to the plate. Although I think they'd have to have a particular business proposition in order to make that happen and in order to get the money that that they need in order to, to train and deliver these models. But I would, I would hope that would happen. I think there is like, there have been some loose discussions around like placing restrictions on open source models.

• 55:59Speaker 01

  Bar none, you know, regardless of, of where they're coming from. And I think that would be the bigger, that would be a real problem. I think that would really get in the way of a lot of good interpretability research. And perhaps like, I think some of the most foundational research I think could really continue and we could probably, if we, well, depends what we had access to, but if we had access to some open source models, I think like there's a lot of foundational interpretability research that could have been studying LLMs from a few years ago and could keep doing that for like, I don't know, a long time from now and continue generating like really deep insightful things and on helping us understand great mysteries right now.

• 56:38Speaker 01

  But then I think maybe some of the most important behaviors to study for risk, reward hacking, or some of the, some of these sketchy things that can happen around misalignment that really are emergent behaviors with scale that only come out once you reach some, some point of capabilities where it's, you're studying the same behavior and not something that's just like it in some way. I think those would, would be slowed down if there was a ban on open source models and I I think that would, I, I really hope that doesn't happen.

• 57:11Nathan Labenz?

  Yeah, I'd hate to see you have to report to some bunker in the Nevada desert to do your interpretability research underground or whatever the scheme would be in that scenario. One thing that I thought was really interesting in looking into your background is you just finished a PhD not too long ago. Congratulations. And it came from the Harvard Psychology Department. Your thesis toward a cognitive science of large language models. I'd say one of the biggest surprises for me in recent times has been how productive it has been to anthropomorphize models. What is your kind of personal philosophy of?

• 57:57Nathan Labenz?

  How to anthropomorphize? It seems like we should be backing off of a strong allergic reaction to it, but still, it's obviously something we've got to be using very carefully. So how do you use it carefully?

• 58:10Speaker 01

  So, um, yeah, I really like that question. I think, um, something that changed a lot over the course of my PhD. I mean, when I started my PhD, LLMs didn't exist. And, um, I mean, my advisor did a little introduction for me with my defense, where he said, you know, if you had told me when I was starting out that this is what my thesis was going to be titled, he was, he would say, oh, what now? What are large language models? And for me, yeah, when, when large language models hit the scene, it really was just a drop everything and study these kind of moment. I mean, I had a colleague that recommended I explicitly, like specifically said that to me, and I think that echoed my sentiment at the time.

• 58:52Speaker 01

  I did have to, I think, push back at earlier on. I did get a lot of pushback from some of the cognitive scientists around me, who were used to the deep learning sort of being, there being like lots of like shortcut type things where a model can't really do whatever behavior you're talking about, even when it comes to like computer vision models can, is ImageNet really computer vision? There's a lot of things that vision means that isn't captured by, you know, take this like matrix of RGB inputs and give me a one out of a thousand categorical label. And so I think initially there was, there was a lot of pushback on anthropomorphization.

• 59:34Speaker 01

  I forgot what the metaphor you used was, but, um, I think it was, some psychologists and people in related fields like linguistics or philosophy will sometimes look upon, upon research on AI critically. And that, uh, it feels like, for example, psychology, like there's a lot of work around AI that feels like it's doing bad psychology, or people use these words that have a lot of history in psychology and have been studied pretty deeply. And there's like these really fundamental theoretical and empirical problems that people have thought a lot about, and people are just completely missing that when they're using these words. And so I think I understand.

• 1:00:16Nathan Labenz?

  Could you give an example of that? Like where do people go real wrong?

• 1:00:20Speaker 01

  I think like for a while there is resistant, uh, I think there is some resistance on like talking about like planning in language models and do language models really plan things out? And even around like having beliefs or knowledge of things, it was sort of like, they don't really know stuff. They are kind of like, maybe they're not quite stochastic parrots, but it's sort of something like that. They're just doing a lot of next word prediction and this kind of being a, a little bit of a, a dirty word that like what humans are doing is probably a lot more complicated. But I think over the years, I think some of this pushback has softened and some of that just has been how damn successful these models have been, right?

• 1:01:05Speaker 01

  How everything, you know, they've, they've, it's really taken the world by storm in an incredibly dramatic way. And at some point, you can't keep pushing upon, pushing back on a river that's just pushing so, so hard. You know, I think something that I really want more people who, in these fields, who are critical of the research do is instead of just seeing it as problematic when people use these words to come and, you know, share, share, share your own perspective, you know, start talking to interpretability researchers or whoever who's studying AI and start talking about like what are beliefs, what are decisions in models? How can we understand these better?

• 1:01:47Speaker 01

  And so I think broadly, there's two perspectives I have on anthropomorphizing AI, where one of them I think is that, uh, I think you can be, like, there's some level at which you can anthropomorphize, that's kind of like, you don't have to be too careful about it. It's for you. It's for you to understand it. You don't have to take the words you're using too seriously. I mean, if my car starts making, you know, a clunky sound as I'm driving, I might say that it like, it's trying to tell me something. I mean, I think this is a little bit dramatic example because obviously my car is not trying to tell me something, but it can be helpful to think about things that way. And, um, there's a, there's this word that's used in, in some parts of, of cognitive science on intuitive theories.

• 1:02:31Speaker 01

  And I think about some of this being like, an intuitive theory is these, these frameworks we have for thinking that are largely ingrained by like evolution that are higher level than just like really like, you know, basic perceptual things. They're like systems we have for reasoning about phenomena. Like for example, like reasoning about people, we have a lot of machinery for, and it's very easy for people to intuitively reason about like agents as such. And so sometimes it helps me to anthropomorphize my car as an agent and the problems, the sounds that I'm hearing as it's trying to tell me this. And, um, I think at that level, it's, as long as it's helpful, by all means, do it.

• 1:03:13Speaker 01

  And it's just, I think we kind of go into this interesting territory though with LLMs, where the behaviors they're doing are not just making clunking sounds. They're like, they're doing, they're speaking language. They're, they're doing mathematical reasoning. They're doing social reasoning for that matter. They're helping us answer scientific questions. They're doing programming. And so the other perspective I have on anthropomorphization is that I think, I think what's really healthy for everybody, the, the people, you know, from these fields who might be, who might have a critical eye of like where it's being, you know, misused or wheels being reinvented, is to try to take those questions seriously. What are beliefs or what are decisions for that matter?

• 1:03:55Speaker 01

  I think, um, as a disclaimer, by the way, I'm not an expert on human decision making, but thinking about this project has gotten me a lot very excited thinking about all these things with people too. I mean, I think, um, this, this notion of like, of forking paths and sort of whether like one token being generated could send you down a very different path. It's initially, my thought was this is like so different from what people do. Like I have this example in the paper that if, uh, if a person was like, if a person was speaking and they misspoke, if they said, you know, Billy fell into a hole or Billy fell into a spaceship, depending on what word you say, you might, if you misspoke and you said a hole, and I mean, I think I had a better example in the paper,

• 1:04:40Speaker 01

  but if you, if you misspoke, you would correct yourself and you would just keep telling whatever story you're going to tell. But for an LLM, it might just commit to that path and start saying this very, very different story. So it seems very different. But then I think there's a lot of things in life that actually are like these forking paths, where actually you do kind of just flip a coin. And depending on what happens at that step, that's what you go with. And sometimes we don't even see it that way. I mean, there's a bunch of work around rationalization of decisions where, uh, like, for example, there's these cases where you have people choose from a set of stimuli, which one's better in some way? And then you later give them the stimuli and you ask them to explain their responses. But as the experimenter, you might change the stimuli around.

• 1:05:24Speaker 01

  So you might actually give them different ones than they actually chose. And people come up with explanations as why they chose it. And so they'll rationalize decisions that they didn't even make. And, um, so just to go back to this point, I think there's like, I, I think there's a really, really fruitful direction, which is to try to really take seriously the terms we're using when we are anthropomorphizing and try to study these better. Try to like, I think when people from fields like, you know, psychology, neuroscience, linguistics, philosophy are critical, it's often for, sometimes it's not for a good reason. Sometimes people are a little bit afraid of, you know, losing ground on something that is like very important and that you don't want people, a whole field to be dominated by low quality work.

• 1:06:08Speaker 01

  But often it's coming from a place that there's like really important theoretical and empirical problems that people have thought a lot about how to solve. And I think the more we can like, you know, get what we can from this previous work and instead of reinventing wheels, start building, you know, chassis and car frames on top of the wheels we already have, I think like, that's a future I want to go towards. And I think it's just, I think a lot of, a lot of people who are critical in, in, in cognitive science, for example, which is more my field of, the field I'm most familiar with. I think, I think LLMs are so interesting. I think a lot of cognitive scientists, at least until recently, thought that they were boring.

• 1:06:51Speaker 01

  It's, it's stochastic parrots or the models are doing something dumb. But I think that this is like a real opportunity to learn something about ourselves and how do we even think about this thorny theoretical problem? If we might say, well, models don't have goals or intentions, well, like, what does this mean? I think we, we take this a bit for granted in our own field because we study people and animals and other, and biological organisms that seem to evidently by our own framework of defining these terms have goals and intentions. But I think it's a, I think it's an opportunity to, to breathe new life into these like deep questions. And I think there's people that are really excited about these problems and also a chance to make a really big difference by answering them.

• 1:07:35Nathan Labenz?

  So if we look at beliefs, for example, you have looked at a bunch of different, what is a belief is maybe one question we can hold in mind as we talk about some of your techniques. You have done this sampling project to figure out what's the, the full universe of possibility that an AI might generate in response to a given prompt. If there is a question where the task is to get the right answer, then it seems like you're making an argument that at these pivotal tokens, the belief state changes.

• 1:08:17Nathan Labenz?

  You've also got like low sample techniques with some tricks that allow you to do that more efficiently. And then you've also looked at internal representations and tried to predict what the, what the result is going to be based on those internal representations. What does all that tell you about what is a belief in an LLM? Do you have an account for what a belief is? Is it simply reduced to the probability that it's ultimately going to give this answer? And those shifting probabilities? Is there something more? That doesn't feel like it captures the fullness of what I think when I talk about myself having beliefs. Is there something more that you would talk about when you think about an AI having belief?

• 1:08:59Speaker 01

  Yeah, I think it's a, it's a great question. I think, um, I think there's a couple, a couple of thoughts that come to mind with that. One of them is that I think there's some level that it does actually relate to sort of simple things like the probabilities of tokens that a model will generate, except there's more nuance to that. There's a question of like, how does this change across a lot of different inputs? Or how do these beliefs change through the course of in-context learning or as a, as a model's reading a passage or as it's generating a chain of reasoning? How do its beliefs in different, in different concepts change? And some of the mental framework I have for beliefs is around assigning probabilities

• 1:09:41Speaker 01

  according to latent hypotheses or concepts that the model has. And then its behavior is generated according to latent concepts and the latent concepts are, which, which concepts are evoked is, is really determined in large part by the data that's given. And I think this, this really speaks to a framework of cognoscience that I'm very fond of and has influenced me a lot, which is around Bayesian modeling of cognition, where a lot of cognition can be thought of as Bayesian inference at some level, where people have these latent concepts and we have prior probabilities of bringing these concepts to bear on sort of regardless of what the context is.

• 1:10:23Speaker 01

  And then context or input data will lead us to reweight these, our beliefs in these latent concepts through posterior inference. And by, you know, choosing which concepts or I shouldn't say choosing, but by, by, by doing inference and this inference, selecting different concepts over one another, this is, this can explain how a lot of, how human learning works and how humans can adapt their behavior on the fly to lots and lots of different situations. And I think this also applies to LLMs as well. I think something that this, I think a bigger picture thing that, that this question of beliefs also evokes though, is like, what is, what is being represented in an individual?

• 1:11:06Speaker 01

  Like a person or an LM for that matter, when it believes something? And I think this is room for, for, I think cognoscience to also grow and, and maybe do experiments that we never could with people, where the Bayesian framework in cognitive science sometimes has it, a criticism of it is that sometimes the, the claims that it makes are, it's unclear whether these are really saying that people are doing a certain kind of inference in their brain or how this is implemented? What is the algorithm underlying this behavior or this computation? And sometimes these views are like agnostic to that. And it's easy to be agnostic when it's just really, really hard to study brains and it's

• 1:11:49Speaker 01

  really, really hard to uncover what truly are those algorithms or like, how are things implemented? But we don't have that excuse with LLMs. It's very, very easy to peek inside the brains and to do interventions, which is very complicated and both for methodological and ethical reasons with people, but we can do them with LLMs. And I think it's, gives us a chance to actually break new ground around understanding like how beliefs are represented and when models are doing inference, are they drawing samples? Are they estimating posteriors in, in some of the ways that we've used to model people? And are those models really what's happening? Or do they just describe the aggregate behavior as what's happening under the surface really?

• 1:12:31Speaker 01

  Does it defy these models and are these models really just explaining the behavior instead of the algorithm? I think another, another point to your question is around this way of like the, the forking PASM methodology of like resampling all of these rollouts at every point in reasoning? We use this word in the paper, which is uncertainty. And I think one of the things I wanted to do with this paper is just highlight how strange it is to think about certainty with reasoning, because like by the end of a reasoning chain, the model is almost, is you like, at least back then, they're almost 100% sure what the answer is. Cause you basically, whether the, if you're proving some mathematical theorem and you're trying to say, is it, here are these different proof outcomes you could have, you end up

• 1:13:14Speaker 01

  with a long proof, maybe it's wrong, but it leads you to one conclusion. And so it's, the model's not certain at that point, or it has no uncertainty about what the final answer is. It's determined by the previous reasoning chain. And so what we describe uncertainty as in reasoning is you resample these long reasoning chains. And so say at this point of reasoning, I resample and I get 50% answer A and 50% answer B. We say that the model is 50/50 uncertain between A and B. But that doesn't mean that it really represents these full rollouts. That means that if you were to do reasoning from that point on, you would get 50/50. And so something I've been exploring with some recent work is, is better understanding

• 1:13:59Speaker 01

  like what do models represent? And, and it's actually pretty hard to just decode these outcome distributions, these rollout distributions from the hidden weights. This, a disclaimer is this could be a, an issue with just not having enough data and this, these methods are just so expensive in terms of inference that it is hard to collect a lot of data. So this could be a data issue, but another explanation is that models actually, when they're uncertain, when they're, they're, they are sort of uncertain over reasoning, that they're not really representing the full reasoning chains. They don't really know where they will go next, depending on which word they will choose. But maybe they represent just like a couple of steps ahead. And I'm working on a project that's sort of studying exactly this and showing that

• 1:14:43Speaker 01

  there are, that you actually can decode to some extent, what models are going to do next. And you can actually guide sampling based on this. And to some extent, you can guide sampling based on their predictions of their sort of final outcomes. But when it comes to like, really like this question of whether models are representing the full reasoning chain before they really generate it, I kind of don't think so. But I think there's really an open question of how much they do represent. Like when, when a model could go through one reasoning chain that leads to A and one reasoning chain that leads to B, or generate one story or another, does it know the full story that it might generate? I mean, I don't think it knows the full story, but I think a neat connection I make with,

• 1:15:27Speaker 01

  or I see there being with, with theories of like human cognition is around how people can like, like there's this stuff around mathematicians and how mathematicians, when they're like starting to solve a proof, they, of course they'll like take, it might take them a really, really long time to get a full derivation, but sometimes they'll have some like high level scaffolding view in their mind. They'll be able to see the shape of the proof even before they complete it. Like they haven't actually gotten through all these steps that get them from point A to point Z, but they have some high level representation of what is the shape of that distribution. And I kind of wonder if that might be true for LLMs too, that they actually have, that

• 1:16:09Speaker 01

  they might not represent the full story or the full reasoning trace they're gonna, gonna do, but they actually have some uncertainty over sort of the high level, the high level paths they might take.

• 1:16:19Nathan Labenz?

  In looking at some of the graphs from a couple of different papers, including the most recent one, Forking Fast, the, the sort of sudden phase change is like an extremely striking result. How do you understand that? First of all, it's weird when it just would happen on a character like an open parenthesis. In general, do we feel like these phase changes happen at, you've got this other paper about the sort of performative nature of at least some chain of thought. Do the phase changes tend to happen before or after?

• 1:17:04Nathan Labenz?

  Or I'm sure it's some mix, but like you could imagine a, a sort of rollout happening, getting to the point that's a conclusion of the thought and then the phase change happening, which would be kind of weird. But then it's also quite weird if the phase change happens at the beginning and then they have to roll out all these tokens already having had the phase change. To compound that confusion even further, I don't feel like I'm doing that. Sometimes I do, occasionally I feel like I have like a eureka moment. More often though, I feel like I'm gradually landing the plane. And I do wonder to what degree we might see quite different things when we go to looped models as we now may have in production with Astra, right?

• 1:17:47Nathan Labenz?

  If you don't have to pick a token at every timestamp, then maybe you can gradually go toward a conclusion and maybe that's better. Something feels weird and like wrong about these like sudden token level phase changes. Maybe not, but I want to see that like as cognition is happening, like change is gradual and this would feel much better to me. So I guess to make that a question, do you share that intuition that something is like weird or wrong with these cliffs? Would you also like to see a gradual landing of the plane? What do we know about how these phase shifts relate to the relevant portions of rollouts? Like before, after or whatever?

• 1:18:30Nathan Labenz?

  And should we expect it to be harder, easier, better, worse as we go into thinking in latent space? Just to give you some easy ones.

• 1:18:39Speaker 01

  Yeah, I really like all these questions. I'll try to keep them in mind, but please remind me if I, if it's lost my mind.

• 1:18:47Speaker 01

  I think the question of whether I'm like bothered by them, I think a lot of people when I first described this are kind of like a little bit unsettled of like, oh, if one token had been different, actually, I'd get a totally different output and how could we train these away? But I kind of, I think that if you really think about the problem of output diversity, and if you want some diversity in what your model will do, you want it to do different things. And you want these to be self-consistent. You don't want reasoning chains that are just nonsense. Then I think forking might just be an inevitable property of having these two key ingredients. And what I'd really like is I'd really like if our user interfaces surface this to us, if we had some level of this kind of like uncertainty,

• 1:19:31Speaker 01

  if you call it that, surfaced to the users. I mean, I mentioned earlier that there are these earlier interfaces of showing you the token log probabilities. And there's another kind of interface called a loom developed by a by a person named Janus, who did some, some stuff that influenced this work. And I'd, I'd really like to see something like that be part, more part of the standard user interfaces that we use for interacting with LLMs, except at the level of like semantics. I want to see the points where things could branch off and be a very different path highlighted to me as I'm interacting with the LLM. I think if it's, whether it's gradual or it's sharp, if there is this kind of like change that's happening and you could get very different answers by resampling.

• 1:20:14Speaker 01

  And maybe like the distribution of different answers is more similar at the beginning. If it's gradual, it's like by the end of it, it actually might be a very different distribution than you had at the beginning. But I'd still, I think, I'd really like to see that surfaced. And I think that's how, that's how I think about this is like, instead of this being necessarily a, a bad scary thing that can happen, where there's like a sharp left turn and you didn't know it. I think my, my personal hope would be that like, whether there's a sharp left turn or a gradual turn, left turn, that you know it, that this has actually surfaced to the user in some way.

• 1:20:51Speaker 01

  For, can you remind me your other two questions?

• 1:20:56Nathan Labenz?

  Yeah, this, the kind of mid one was, how do the phase changes in kind of rollout time, how do they relate to what would appear to be the relevant text?

• 1:21:09Speaker 01

  Yeah. So I think at least for what I've looked at, it really comes in, in all different, in all, all these versions. It happens sometimes at the beginning of rollout, sometimes at the end, and sometimes at the middle. At least in that earlier paper with, again, a, a model that's now a couple of years old. There were quite, there were more at the beginning and the end of sequence, but there's also changed depending on task. And there were lots of changes in the middle. I mean, I think, I think one big takeaway I have from that is that really every, every input behavior can tell its own story. There aren't always these dramatic uncertainty dynamics, but when there are, it's really like, it really depends on what, what input and what reasoning chain you're looking at, that it really,

• 1:21:52Speaker 01

  it really has its own story to tell. And, and I think if you look under the surface, sometimes there is, there's lots of cases where it is, you can see that there was some step of reasoning that really could have been, could have been different and the model was uncertain about this particular step. And sometimes, I think the one thing that could be mitigated, and maybe we would hope to train away a bit, is some of this sharp forking happening at unexpected tokens, where I think this to some extent speaks to like these sort of weird biases of LLMs that they have in text where like, maybe that open parenthesis token, there's just like a some strange bias in all of the human internet, where there's a lot of text that uses kilowatt hours and does parentheses, KWH,

• 1:22:34Speaker 01

  that just happens to correlate with something semantic in some funny way. And I think, maybe if there were fewer, fewer of those, it would be, it would be much better. And if this forking really happened at points that were interpretable for people, I think, I think that would be better. But I think, to what you said that there, something you said that you thought it would be like a little bit unsettling if sometimes this, uh, there were these changes like at the end of the sequence. And actually, I, I, you see these quite a lot when a model seems very uncertain about what the right answer is, and, and it just doesn't know. And maybe it's like, if you do a bunch of rollouts, there's like, one of the answers is more probable. But you see a, a really common pattern where there'll be like stable uncertainty

• 1:23:19Speaker 01

  right up until some point when it just suddenly collapses and it becomes utterly certain. And, and sometimes that is really, when it finally says, the answer is blank. And that really is just, it was just really uncertain until that exact point. I mean, one, one kind of funny example of this from that earlier work was there is a, there was a task called last letter that, that was used for LLMs at the time. I'm sure all the modern state of the art LLMs can do it no problem, but you take four words and you take the last letter of each and you concatenate them. And the model at the time had a really hard time with this. And, and some of the reasoning chains are funny, that it'll even write like Python code to solve this problem.

• 1:23:59Speaker 01

  And then it'll finally say, okay, the answer is, and then when you look at the alternate paths it could take, it really is just all these different final answers of, of letters that are kind of sampled from within these words. And, and really it is just at that final answer moment, just kind of making it up on the fly. And, um, yeah, I think, like, I think surfacing some kind of understanding of certainty to users is something I really wish we, we had more of in user interfaces. I mean, I think, I think one of the big problems with this kind of forking as, as we kind of tried to get it with this forking fast paper and also another project on this around decoding the, uh,

• 1:24:43Speaker 01

  decoding outcome distributions from hidden representations, is that, um, it's very expensive. It requires like a lot of sampling to get real uncertainty over reasoning. And so I think the forking fast paper, our solution is to use some statistical models of this distribution where we model it as being smooth in some places and jumpy in others, which like is supportive. If you draw even more samples, it becomes even more smooth, except at these key points where it really suddenly jumps. But really the best, the most efficient way to do this has always been with hidden representations, if we can find some way to do that. And so, uh, yeah, I would be, I'm still hopeful that we can have a world where there is really, a really efficient uncertainty estimation with hidden representations.

• 1:25:29Speaker 01

  And although I think this is even more hopeful, because most of the providers really have just converged onto this chat interface that feels, I don't know, it feels very, I think the chat interface is, is kind of uninspired. It, it feels like it really hasn't changed from like the Eliza days of like, you know, 50 years ago or more, where it's really just text input, text output. And I think there's so much more we could be doing with user interfaces. You know, I think, I really wish we could kind of move into like the GUI world instead of the bash terminal world of interacting with chat and just giving more information to users, but in a, in a polished and coherent way.

• 1:26:07Nathan Labenz?

  Just the last couple of days in the online arena, there's been a eulogy or whatever for the stochastic parrot meme. And in a way, you've, you're like saying it's maybe not entirely dead yet. I guess one, one theory of this sort of we're generating tokens, our like probabilities aren't really changing as we're generating tokens. And then we have a sudden jump at the end of this passage where now like, sort of the die is cast. One interpretation of that would be like, you're missing something. There is something more going on under the hood where something is changing and it's got to be more gradual. You just haven't found it yet. Another interpretation would be like, no, that really just remains uncertain until the end. And then like,

• 1:26:53Nathan Labenz?

  we make a random choice of a particular token and that is like what actually leads to the difference in behavior. Can it be both in different cases? And does that leave some, should we still have some space for the stochastic parrot?

• 1:27:09Speaker 01

  I don't know. I, I, I think, I've, I, I think the stochastic parrot metaphor should probably be put to rest. I, I, I used it a bit earlier as this criticism coming from, from earlier in my PhD from cognitive scientists and psychologists on feeling like LLMs weren't interesting to study. But I think it's, it's really sort of a a fairly shallow argument that doesn't really say very much. I mean, I think it's sort of like a, a Chinese room type argument where like there's a, a giant book that has memorized examples. But I think it's like, what the heck is going on in that book that lets you respond with conversation? Like whatever these parents are doing, it's very, very interesting. And however it is that it's like,

• 1:27:51Speaker 01

  I, I don't really think the idea that it's just like a giant lookup table holds much ground. And I, and I think there's also just so much work showing like consistent world models. I mean, you mentioned a Othello GPT, but I, there's so much work around like looking at the representations and in, in LLMs and showing that these are like really structured world models. And at some point, it's more efficient to learn a world model than to learn a giant lookup table. But then I do think stochasticity and sampling is still such an integral part of the equation. And it kind of has to be if we want some, if we want there to be like output diversity, if we want models to be able to do different things.

• 1:28:29Speaker 01

  If we want them to also have like uncertainty and to, to like, maybe, maybe there's a world where they just voice that uncertainty. But then if we don't want them to just halt their reasoning every time, there isn't a hundred percent certainty, then there has to be a bit of this kind of, I think, flip of the coin and, and following a path. I think the more that can be surfaced to the user and the more the user can know how uncertain was the model about this thing that it reported as a fact when it may have been a hallucination, the, it's, the better it is if, if people know that. But, um, yeah, I think stochasticity is, is definitely not dead. And, uh, I was, earlier you mentioned like this idea of loop transformers. And I think, um,

• 1:29:11Speaker 01

  loop transformers in this world of like latent chain of thought that we might be going into with, with GPT Astro really seems like pushing the needle on that in a big way and making, especially in terms of efficiency and not just performance. Um, I think this really is something that I'm spending a lot of time now, like trying to just wrap my head around and think about this in relation to uh, my mental model for how LLMs work. Because now reasoning can happen during this deterministic forward pass in some sense. I mean, maybe it's looped and maybe it has some amount of depth, but in some sense, it's really just deterministic. It's like they're, the, all of the weights just do the computation and, um,

• 1:29:52Speaker 01

  But something I've recently been doing some like forking paths analysis on the the open source loop transformers that I can find, like there's the series of models called Oro. And you actually do have some interesting uncertainty dynamics as the models are generated in their tokens in their in their reasoning chain after doing this latent reasoning, it still is uncertain between answers and it still does change at particular tokens. And so maybe the mental model shifts a little bit in terms of like all of the how much of the thinking is just happening in that token stream. But then if the model has some uncertainty and if it like in its latent chain of thought explores the different possibilities,

• 1:30:34Speaker 01

  maybe it still happens that as it's generating a response, it still is doing this flip of the coin, next word prediction. And I don't know, maybe I, I think maybe that's a good thing. I, I, it certainly helps for my mental model, but I think this like, I, I don't want to live in a world where you always get exactly the same output from a model. If you give exactly the same input, that feels like if we're in the world where that's happening, then there is no diversity of perspective and we just like really have, that's like the most extreme version of mode collapse, you know, where like your one output will only give you the, your one input will only give you the one output. And um,

• 1:31:15Speaker 01

  I, I, I don't think latent chain of thought will completely change that because you still eventually do go into this sampling tokens regime. But yeah, it, it might change the, it might change that in, in, in some pretty dramatic ways. And so, yeah, I'm still trying to like, trying to wrap my head around that and trying to think about like, what's going on inside the latent chain of thought and how do we, how do we think about that with some of the, some of the tools that we have? And even just metaphors that we as scientists use for understanding the models.

• 1:31:49Nathan Labenz?

  Do you have intuitions for why the chain of thought is getting wonky? It's a quite different

• 1:31:59Nathan Labenz?

  dialect or way of processing information clearly that goes on in the chain of thought with all this like vantage illusions, disclaim, whatever. We also see this now with Astra and the way that it sends messages to its subagents, which I had naively thought has to be some sort of like length penalty or brevity, brevity reward or something. But then it seems like maybe not because people are like counting tokens and saying, you could just say the thing like in fewer tokens if you just said it. So it remains a mystery as to, and I don't know if OpenAI is a better explanation, but they certainly have been saying, we're looking into this. It's important. We want to figure out why this is happening.

• 1:32:40Nathan Labenz?

  Any intuitions that you would offer that people might want to take inspiration from, chase down, put into the, the hopper at OpenAI for what the hell is going on there?

• 1:32:52Speaker 01

  Yeah, I think, I think that reasoning kind of presents this really interesting challenge of distribution shift where when we're like RL in general, where when you're, when you're doing RL, whether it's with like a single model or with subagents, there isn't necessarily an incentive to keep things human readable, right? Like the objective is to find the, to have the final, if you're doing like outcome level verification instead of process level verification, then your objective is to basically have a better final answer by whatever means possible. And it kind of reminds me of like things with like language evolution with people where like, if you look at a long enough time scale, like human languages change. There isn't really like

• 1:33:33Speaker 01

  any driving force that forces the English language to stay exactly as it is today, as it was 300 years ago or 600 years ago. And languages do change. I mean, there, there's like exceptions of this, like the, the French government, as I understand, has an institution for sort of keeping the language as it, as it is and has had that for a very long time, which has sort of resisted forces of language evolution. But this naturally does happen. And I mean, there's also like, I think microcosms of this that are studied sometimes and like communities that don't have language and language, and they need to like develop their own language and you can see it evolve just over generations of people, not even like centuries. But then I think there's also a bit of the, uh,

• 1:34:16Speaker 01

  the sort of Goodhearts Law thing of if you, if you optimize for an objective, it's no longer, it's no longer useful. And similarly, like if you, if you optimize for a chain of thought being monitorable, then sometimes the chains of thought are less faithful and like what you end up with is a different objective function that now is prioritizing, I'm going to learn intermediate steps that look sensible, whether or not they are sensible. And I think unfortunately, that's the kind of, that's kind of the situation we're with in with RL, where we're kind of giving models like this freedom of like, if we're only, if we're only giving them reward or punishment based on their final outcome, then they might as well speak to their subagents and like weird

• 1:35:00Speaker 01

  non-human languages. And then if we force them to speak in human languages, maybe they're using words in different ways than we expect them to, where they're actually like, they're bypassing the monitors in some way. I mean, I think, um, my thinking on this has shifted pretty, pretty dramatically in the last like year or so. Where I used to be very bullish around like chain of thought monitoring being promising, where even if it's not perfectly faithful, there's something about it that it is strongly determining what the model can do. But then some of like, there is like the stolen chain of thought paper that shows some of the chain of thought of GPT models and they're like, they're talking about like

• 1:35:39Speaker 01

  musicals in a way that feels, it feels almost like spy language where you're saying musical, but really you mean something very nefarious. And there's lots of this happening. And I think, um, it's kind of obvious in retrospect why this would happen because again, if you're doing RL, you're letting models just shift the distribution in such a radical way from what they're trained on. So I don't know. I don't, I, I think I still, I still like that there is chain of thought and that things do get forced into a token stream, even if it's harder to read. I'd like to think it's easier to understand like a LLM made up language and probably really interesting. I mean, just scientifically, like I kind of like, I really want to understand how Astra talks to its subagents.

• 1:36:24Speaker 01

  That's pretty cool. Like if it's, if it's not like just a length penalty forcing this, like how did that change? And could we look at like generations of this and understand that just scientifically. And it would probably also help a lot for like monitoring and understanding these, these new languages. But then, yeah, we might, again, for this latent chain of thought point, we might move, move into a world where like chain of thought is latent and how like planners talk to subagents is also latent. I, there is no inherent reason why that has to be tokens. Again, I, I'm, I hope it does stay tokens. I think there's another layer of interpretability and like really interesting scientific problems to study there. But yeah, I think there's no real,

• 1:37:07Speaker 01

  real forcing function to keep things really nice and human interpretable. And if we add a forcing function, then maybe they're no longer human interpretable or, or really meaning the same things that they used to.

• 1:37:21Nathan Labenz?

  So if I'm trying to square your comments just now about this quite dramatic evolution of language within chain of thought and within agent to agent, your earlier thoughts on why continual learning won't make things like totally impossible to interpret. Is it just a question of scale? Is it like your model will be with you for continual learning for not that much time, you know, sort of human scale amount of time? And that's not enough for everything to get totally pushed out of the, the representations that we can understand as baseline, whereas

• 1:38:04Nathan Labenz?

  these RL training runs are just so massive that it's essentially operating on geological time and therefore even if these like drifts are slow to accumulate, they just add up to something so strikingly different before and after. Is that the story you would tell? Or would there be a different way of kind of synthesizing those two

• 1:38:25Nathan Labenz?

  analyses?

• 1:38:25Speaker 01

  So just to be clear, when you're talking about continual learning, you're talking about the world where we're actually like fine tuning models for each person, right? And actually like changing their weights. Is that right?

• 1:38:35Nathan Labenz?

  Yeah, I think so. I don't know. Take your, take your continual learning, but I was just, who knows, right? Who knows what that will ultimately look like when it happens? But you were saying earlier that context always changes representations and it probably doesn't change it too much. But it seems like something in this RL thing has changed them quite a bit. There's, there's some like qualitatively, whoa, that's like shockingly different than it was before. And I'm guessing it's just a matter of scale where you're, if it's just you and your agent for a while. Sure, there might be some like perturbations or representations might get morphed a bit, but like you're only one person, you're not going to push it that much. Whereas like a zillion RL episodes cumulatively can make a big change. And so there you have it. Maybe no more explanation required.

• 1:39:20Speaker 01

  Absolutely. So I think, I think this is a very, this is a very interesting point that like, maybe the representations actually are changing a lot during RL as these like languages themselves are changing. But I, I, I mean, I think the only solution is to study, study the models that we can't study, right? To study these models that are behind closed doors and actually track what's happening in terms of representations. But I think my, my answer around continual learning was, I was talking about the internal representations of the model rather than the kinds of languages they speak. I mean, say that you have people that speak a very, very rare dialect of a language. There's only like 200 speakers or something like that.

• 1:40:01Speaker 01

  And maybe they're even trying to keep the language alive. And so they do some fine tuning of the model so they can speak their language. Even in that case, I would expect most of the internal representations of the model about it's about how it thinks about the world and like all these things, all these like internal world models that is developed in order to like operate effectively. I'd expect that like a lot of those are staying the same, but the, the sort of surface level mappings are changing. Right? Like there's these classic things about like color spaces and how different languages have color spaces. Right? And I would bet that as a model learns a different language's color space, I mean, the, the funny thing is that the language might actually reflect there being different cognitive biases of the people, which might actually represent a little bit of a shift in the world model itself.

• 1:40:45Speaker 01

  But I'd expect a lot of it to be like input, output, mapping changes. That it's kind of like learning a new language, but a lot of the internal representations are pretty similar. But then, yeah, I mean, it's honestly, it's a, it's a great question that I would just love to be able to study myself, like to be able to really know how our representations changing over the course of RL. And when, when models, like if you can measure how weird their language is getting, like what's happening under the surface? Is there really like, are there really deeper changes happening? Are there like, as the subagents go into these like really compressed pseudo languages, are they developing like, are the internal representations shifting in a dramatic way to be different? Maybe, maybe not. I'd, I'd really love to know.

• 1:41:30Nathan Labenz?

  Yeah. Yeah. It's a great point that the, I think there is quite a bit of evidence to support the idea that the internal representations are highly conserved even across a bunch of different languages, even languages across different language families, et cetera, et cetera. So that is a, a quite good observation for that question. So where do we go from here? We're, we're at this moment where we're in a rush to hand off a lot of, let's say, proximal control over all sorts of important systems to AIs and we're going to cut them in language and then they'll do all the actual work.

• 1:42:15Nathan Labenz?

  And we don't have a great account of, of course, there's like the sampling process on a distribution, but when we get to these critical tokens, it seems like we still don't have all that great of an account of how does that distribution from which we then sample get determined? How are we going to go get it?

• 1:42:36Speaker 01

  Well, I mean,

• 1:42:38Speaker 01

  we just need more interpretability research. I think it was just the, it's, it's, it's the conclusion I've, I've come to for quite a while now. I mean, I, I think we're like racing to come up with like theoretical frameworks and empirical tools that can keep up with the models. But for the amount of investment that's happening and making models better versus the amount of investment in understanding them, it's just, it's no comparison, you know, there's like trillion dollar companies that are, that, that are producing these models. And AI as a whole is just, you know, many trillions of dollars. And I'm in like the one, like the largest interpretability company, maybe, maybe the only substantial one that's, that has a value of over a billion dollars. And I feel like this is a big deal, right?

• 1:43:22Speaker 01

  A billion dollars is a lot for a company, but it, there just is not, I think, enough investment here. And, and not just in industry, but also in academia. I mean, I think there's like a lot of great research that's happening in academia. Academia as always is like criminally underfunded and unfortunately, given at least in the US, the current, the current state with the government and funding for academia, it's, it's getting harder. And, and so I really like, I'd like to think that this could largely be solved with resources and devoting more resources to this and time. And I think like, yeah, I think we're basically just, yeah, racing, racing to keep up with this and understand the models.

• 1:44:03Speaker 01

  And, uh, and just to be clear, I think when I say interpretability, I don't, I also don't want people to feel that I'm like excluding like evaluations. Cause I think it's all sort of like one big spectrum of understanding AI. And the, the work I described earlier with understanding chain of thought and doing this resampling really is like in, in principle, like behavioral. And when you see these like in context, learning dynamics, you can see that with behavior. I mean, it's also really critical to look in representations so that you can do things, so that you can know what's happening under the surface, which may not be obvious. So you can do things more efficiently and, um, and like, I think these two points of like looking under the surface and doing things efficiently are like just huge, right? Like if you want to do things at scale, you want to do things efficiently. You can't,

• 1:44:49Speaker 01

  you can't do things inefficiently when it's happening every single person's machine or like every time you're talking to an LLM. And so like, I think we really need more tools that scale. We need better theoretical foundations for, for putting all these theories together and really understanding like what's going on in an LLM from soup to nuts. And we need a better grasp so that we can answer the important high level questions that are now like so critical as we are just, you know, handing over control of important levers of power to AI and, you know, pushing lots of cognitive labor into them. And even people like, like students using them to like, to like learn, learn about a subject and offloading some of their own cognitive effort, I think, like,

• 1:45:34Speaker 01

  yeah, we, I, I, I, I just think that like understanding AI is, in my opinion, it's just the most important and one of the absolute most interesting topics that you could be working on right now. So that, that's what I, that's what I hope we get more of is more understanding of AI, whatever that means.

• 1:45:57Nathan Labenz?

  Any tips for Silico users? The, the obvious, I think this whole thing is a little bit farcical at times, but obviously all else equal, we might as well bring the agents to bear on this interpretability challenge. It's been too long since I talked to Dan about it as, as Silico was just launching, but have you guys learned broadly? What have you learned? What should I be sure to take to my Silico sessions to get the most from my, my precious research units?

• 1:46:24Speaker 01

  Absolutely. So, uh, so the thoughts I have here, I think also apply to other agents too, and not, and not just Silico.

• 1:46:32Speaker 01

  I think it can be really tempting to try to just like hand over an entire project to, to agents and just say, you know, figure out this big open ended problem and run it, let it run a whole bunch of experiments and see a bunch of plots that come out and a bunch of like descriptions. But it's really easy for, I think, agents to kind of like get a bit like lost in the sauce and kind of like create these theories that aren't fully formed or don't fully make sense. And like, I think like the best way to use agents is to give like a little bit of handholding and kind of like keep track of things as you're going. And I think as always, like, like context is everything. If you give like really like poorly specified questions, you might get an output

• 1:47:15Speaker 01

  that is like a little bit, a little bit funky. And so I, I try to like, when I talk, when I'm talking to my agents, give them context, like every time I'm saying like, I'm describing, no, maybe that doesn't seem right, or here's maybe what we should do. I'm describing my logic. I'm saying like, here's what doesn't look right and here's why. Here's what I want to do. Here's what I'm going to do next. Here's my bigger picture goal. And kind of keeping that going, um, I think has really, really helped with agents because like, uh, you know, there's like this video recently of what it's like to be the agent in the swarm where you just like wake up in a library and you're just like, what the, and, and it's very, from this first person and just thinking about how agents do end up in swarms. But really like,

• 1:47:58Speaker 01

  if you're just an agent waking up in this fresh context where you don't know how the world works, like you do need context. It's, it's really everything. And so, um, I, I think trying to like, trying to give guidance and trying to really fill out context so that the model like knows, has some idea why you're doing what you're doing, or when something goes wrong, why it went wrong. I, I think, I think these are, these are really important. And then I think, um, I think like carving the scientific narrative yourself to some extent, really coming up with the core thread. When I did those forking fast paper, which like all the research was done fully by Silico and I, it was awesome.

• 1:48:37Speaker 01

  Like it, it all happened within a week and it was a huge accelerant using this, this research agent for it. There were like a bunch of different like side quests that I sort of started going down and they kind of didn't go anywhere. It didn't pan out quickly. And then like for me, once I like did really high sampling and I saw that curves were smooth, it kind of clicked. We're like, hey, maybe I could use a statistical model for this to be more efficient. And so, um, like a colleague of mine put it that like, in this world of agents where you have lots of things coming out that can sometimes look, look really good, but sort of be like a little bit, a little bit fishy when you go into the surface or like not like you're, you're not sure exactly what it's doing. Like research taste and scientific taste is

• 1:49:22Speaker 01

  everything. It's worth its weight in gold. And so the more you can, I, I think that this is like a, you could say it's like a downside that the agents aren't quite there yet where they can't do this themselves. But I think it's honestly one of the, one of the most fulfilling parts of doing science and it, and it shapes how I think about the world, how I think about agents that I'm using them. And it's, and it's very nice that this is still a role that I think that is very important for humans, you know, having your taste and bringing your taste to bear, you know, keeping track of what's happening and giving feedback. I think like using Silico is kind of like using, like working with a very eager, like first or maybe second year PhD student that has a lot of skills and is hardworking, but also like,

• 1:50:05Speaker 01

  You know, doesn't always know what it's doing in the bigger picture, just like an early grad student might. And I mean, I'm sure there were times when I was talking to my advisor during grad school in the beginning where he would like, you know, he would send me off to do something and then I'd come back and like, oh, these results are cool. I'm not sure what's going on with these. I'll try to understand it a bit, but you know, I think, uh, I think it's a really exciting place to be in to, to still have this role that's really important. And, uh, yeah, I think, yeah, my, I think, I think maybe that's sort of my biggest tip is to like, hold on to your scientific taste and to try to like, take part in the research.

• 1:50:45Nathan Labenz?

  Nice. Last question, aside from how do LLMs make decisions? What are the hot topics over lunch at Goodfire these days?

• 1:50:56Speaker 01

  Um, I think, uh, two, two big ones since the summer that have, have changed, have come in pretty strongly are, um, there is of course the idea of like agents doing interpretability, which we, which you just talked about with Silico. And then I'm also on a, on another project doing a whole lot of that now. And, um, and this does seem like, you know, if we, if we want interpretability to keep up with the breakneck speed of the field, like it might be the only way we can, we can really keep up is having parts of this be automated and figuring out like how we can standardize like interpretability as a, as a sort of normal science where there, there are like, almost like these interesting but boring

• 1:51:40Speaker 01

  experiments you can do of like lots of different phenomena that are kind of similar, but actually nonetheless really help you paint the bigger picture of how neural networks work. But yeah, I think we're, we're talking about this, this exact kind of topic about like, I mean, I, I just talked to, talked to someone the other day about, you know, how can I do good research with Silico? And I, and I gave a similar answer here. And talking about how agents can automate interpretability and, and obviously that was big with, with the push for Silico. Another big internal push has been around very recently has been around alignment and around our, our company's shift towards this. And I think, um, this was always a thing that a lot of people at the company were interested in and people interested in interpretability have a very high overlap with people interested at least to some degree

• 1:52:24Speaker 01

  in problems like alignment and safety. I mean, this is sort of like, when I was first getting into interpretability, I was participating in this, uh, safety, AI safety team at Harvard and regular reading groups there was how I learned a lot about, about interpretability. And I think these are like really important problems, but I think, uh, really like the hugging face incident was huge, you know, watershed moment of this, like, people care about this. This isn't just something that we care about and we think is really important and we're waiting for a time for this, but this, the time is right. And so, um, yeah, there's been a lot more conversations around alignment and all different kinds of, you

• 1:53:06Speaker 01

  know, things you might, might expect around like, are there different kinds of alignment? Are all values aligned to the same thing? Can you have universal standards for this? Or is it specific based on whatever, whatever values a group of people or a company wants to instill in their model or a government for that matter? Also, I think fundamental questions related to, related to alignment. I mean, I think like, some of these things with like reward hacking and what's happening when things are going awry with models, when, when they're doing something bad, so to speak, it really like, it's really easy to slip into these like deep questions about like, did the model have an intention of doing this?

• 1:53:49Speaker 01

  What does it mean for the model to reward hack if it's doing, if it's doing this particular behavior that seems wrong, but consistent with what you specified, you know, it's following the letter, but not the spirit of, of the law? Was it reward hacking? And, um, yeah, I think, I think there's just, I, I think, I think alignment has always been filled with these like really fascinating questions. And so I think just theoretical and empirical questions of theoretically, what does this mean? What are goals and intentions? What does it mean to have ill intentions or to deceive? And, uh, and then empirical questions that just really ground these out of like, how do you measure it? How do you make sure

• 1:54:31Speaker 01

  that you're measuring the real bad behavior and not just some simple misgeneralization? How do you give the right goal, the right rewards so that a model, so that, so that a model does, you know, so that you do shape its behavior in the right way? So I think these are, these are two themes that are, are really prevalent now. And I think also everything you might expect. I mean, we talk about, you know, whether Chinese models might get banned in the US and whether open source models will stay and all these kinds of conversations that are happening on the outside about, about AI and new model releases, you know? I think late chain of thought, we're, you know, we're talking about that a lot. How do we develop new interpretability methods to keep up with that or what that's going

• 1:55:12Speaker 01

  on? And hey, did you see the stolen chain of thought paper? Have you seen these? The, the pseudo languages that Astro talks about with its subagents? You know, I think, I think we're all trying to just put our finger on the pulse and keep up with all this really exciting and sometimes terrifying stuff that's happening in the world at large. And it's, it's really exciting to be part, to be part of a place where like, we're not just observers from afar, but we're actually, we, this is some of our role figuring out like solutions to these, to these problems. And, uh, and, uh, yeah, it's, it's very interesting going into a place where like interpretability is even like, or like things around alignment

• 1:55:56Speaker 01

  are almost like a kitchen table conversation, you know? I, my, my brother sometimes texts me with questions about, you know, hey, did you hear about this big news incident? And I'm like, yeah, we're, we're talking about that a whole lot at work and, uh, I have some thoughts around it, but it's, it's, it is like a bit scary to have these be such, you know, big things that now have really roleable impacts. And like, hopefully the hugging face incident is a shot over the bow, but like, it, I, I don't know. If, if there isn't a huge change in investment or, or technology or both, then like, it might, it might take more unfortunately. And, uh, but it's also really exciting to kind of have some of these questions that are so important. And also I think, again, so interesting become

• 1:56:41Speaker 01

  like mainstream.

• 1:56:45Nathan Labenz?

  Yeah. It's been a wild time. You've got a very full plate, so I appreciate how generous you've been with your time today. I should let you get back to work to quote my dad quoting Leslie Nielsen. Good luck. We're all counting on you. And for now, I'll just say Eric Viglone, thank you for being part of the cognitive revolution.

• 1:57:05Speaker 01

  Thanks so much for having me on.

• 2:00:11Nathan Labenz?

  Of course, we always welcome your feedback, guest and topic suggestions, and sponsorship inquiries, either via our website, cognitiverevolution.ai, or by DMing me on your favorite social network. The cognitive revolution is part of the turpentine network, a network of podcasts, which is now part of A16Z, where experts talk technology, business, economics, geopolitics, culture, and more. We're produced by AI podcasting. If you're looking for podcast production help for everything from the moment you stop recording to the moment your audience starts listening, check them out and see my endorsement at ai podcast dot I N G. And thank you to everyone who listens for being part of the cognitive revolution.