# Frontier AI Models Fail at Automated Research | Qasim Wani

Markdown copy of https://realalpha.app/v/fH7CFf1wrpA/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Qasim Wani argues recursive self-improvement depends on high-quality data and research environments, not just stronger models.

- Show: MTS
- Hosts: [Theo Jaffee](https://realalpha.app/people/theo-jaffee/) & [Sophia Dew](https://realalpha.app/people/sophia-dew/)
- Guest: [Qasim Wani](https://realalpha.app/people/qasim-wani/) is co-founder of Extrapolate and former xAI researcher.
- Published: Nov 6, 2026
- Length: 24 min

## The best minutes

- [1:36](https://realalpha.app/v/fH7CFf1wrpA/?t=96) [Recursive self-improvement means inventing a new lab](https://realalpha.app/s/recursive-self-improvement-means-inventing-a-new-lab/) (6 min)
- [10:20](https://realalpha.app/v/fH7CFf1wrpA/?t=620) [High-quality data is the bottleneck for recursive improvement](https://realalpha.app/s/high-quality-data-is-the-bottleneck-for-recursive/) (4 min)
- [18:53](https://realalpha.app/v/fH7CFf1wrpA/?t=1133) [Models still struggle to diagnose a failed training run](https://realalpha.app/s/models-still-struggle-to-diagnose-a-failed-training-run/) (3 min)

## Also worth watching

- [14:28](https://realalpha.app/v/fH7CFf1wrpA/?t=868) [Synthetic coding tasks can leak answers through commit patches](https://realalpha.app/s/synthetic-coding-tasks-can-leak-answers-through-commit/) (2 min)

Skip the opening setup, the Lovable read, the discussion of experiment selection and RSI risks, and the closing sponsor block.

## Summary

Qasim Wani is co-founder of Extrapolate, which builds reinforcement-learning environments where AI agents learn to improve through frontier research. He previously researched at xAI, where he started RSI efforts, and contributed to Semi-Analysis’s GDPX evaluation.

Wani defines recursive self-improvement as an agent’s ability to invent a new lab, not simply automate parts of model training. He says frontier models still struggle to choose valuable experiments, with GDPX showing average accuracy or mean reward of no more than 24%. He argues that progress depends on high-quality data and research environments, pointing to weak synthetic tasks and leaked code patches as evidence of reward-hacking risks. Wani supports automating some lab work but says data-quality auditing remains far from solved.

### Minute by minute

- [0:58](https://realalpha.app/v/fH7CFf1wrpA/?t=58) **RSI means more than automating training.** Asked whether a reported acceleration in Anthropic’s capabilities scores and R&D uplift signals RSI, Wani says labs define RSI differently. He describes Anthropic’s reported focus as automating pre-training research, but argues that self-improvement involves more than model training.
- [3:02](https://realalpha.app/v/fH7CFf1wrpA/?t=182) **Research is iterative across many benchmarks.** Wani says lab research is iterative and researchers spend much of their time examining data. He argues that benchmarks focused on a single task, such as optimizing one model for one benchmark, miss how labs hill-climb across 20 benchmarks at once.
- [5:34](https://realalpha.app/v/fH7CFf1wrpA/?t=334) **Wani defines RSI as inventing a lab.** Wani defines RSI as an agent or swarm of agents being able to invent a new lab. He distinguishes that from self-improvement that merely shortens model-development cycles, and says GDPX measures capabilities he sees as important to achieving superintelligence.
- [7:38](https://realalpha.app/v/fH7CFf1wrpA/?t=458) **Models struggle to prioritize experiments.** Wani says frontier models scored no more than 24% in average accuracy or mean reward on GDPX. He says models such as Astra, Carbon and Opus do not reliably identify which experiments are most valuable, estimate how long they should take, or use existing logs and research to guide the next bet.
- [9:49](https://realalpha.app/v/fH7CFf1wrpA/?t=589) **Wani sees data as the RSI bottleneck.** Asked whether models need full knowledge of a lab to improve it, Wani says his team’s work outside a frontier lab is a deliberate bet. Drawing on his experience at xAI, he argues that RSI is fundamentally a data problem, and says data vendors often lack the expertise to produce frontier-grade samples.
- [12:55](https://realalpha.app/v/fH7CFf1wrpA/?t=775) **Data and environments remain weak.** Wani says data is undervalued: he puts the total data industry at no more than $150 billion, compared with model and chip companies worth more than a trillion dollars. He cites a benchmark where the best model passed 49 to 50% of environment-creation tasks, and says agents building synthetic coding tasks accidentally leaked deleted code patches; Xiaomi’s paper released about 21,000 synthetic environments.
- [16:35](https://realalpha.app/v/fH7CFf1wrpA/?t=995) **Wani sees safety work in better data.** Wani says he approaches RSI risks as an engineer trying to solve them, alongside the need for better policy and governance. He warns that poor data can create reward-hacking or other harmful behaviors, and says Extrapolate is building safe, aligned datasets for RSI capabilities.
- [18:53](https://realalpha.app/v/fH7CFf1wrpA/?t=1133) **Data auditing is not automated.** Wani says some lab functions, including starting and monitoring training runs, have been automated, but auditing data quality remains unsolved. In a GDPX task, models received 20 million tokens from about 1,000 rollouts and five hours to diagnose a reward regression; their answers were usually different and incorrect, with the top score around 24%.

## Transcript

MTS, hosted by Theo Jaffee and Sophia Dew, with Qasim Wani.

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[0:00](https://realalpha.app/v/fH7CFf1wrpA/?t=0) For GDPX, we observed that the average accuracy or the mean reward for all the frontier models was not more than 24 percent, which means that there's still a lot of room to grow for these models to improve on. Part of the reason why these models are not good at it, and we sort of talk about it in our blog post within the GDPX report. one of the big problems with models like Astra or Carbon and Opus is that they don't know what experiments to prioritize. We're live with Qasim Wani, who's the founder of Extrapolate, which builds RL environments, or co-founder of Extrapolate, which builds RL environments where AI agents learn to improve themselves by doing frontier research. And it's also contributed to GDPX, which is Semi-Analysis's new eval, which seems really cool and interesting.

[0:57](https://realalpha.app/v/fH7CFf1wrpA/?t=57) So Kassim, welcome to MTS. Thanks for having me. So where should we start? We were just talking before we went on about the new meter report that says that basically Since October of last year. Yeah. Since October of last year, the rate at which the Anthropic Epoch Capabilities Index benchmark score has been increasing is about double what it was previously and they have other evidence that's non-public uh in addition and then all this combined uh leads them to believe that anthropics models are providing like substantial r d uplift already so do you have any takes on this like if if this this line is true like does this mean we're entering rsi um it's one of those like if this then that kind of situations um i think what's interesting to zoom out a little bit is to define rsi more concretely uh something that we're talking about right before the show is different labs have a different perspective on rsi and it's it's really interesting how a close friend of mine he he works at google and they claim that they're also working on rsi but his team specifically does not work on model training so it's like what even is is rsi at this point what does rsi um it's it's like it's the thing that's in vogue um the way that i would

[2:27](https://realalpha.app/v/fH7CFf1wrpA/?t=147) **Qasim Wani:** say i think about rsi is can an agent invent the next version of itself um you know anthropic in

[2:34](https://realalpha.app/v/fH7CFf1wrpA/?t=154) particular their interpretation is from at least what i've heard so internally things could be a little different but from what i've heard their whole interpretation of rsi is uh how good is model how good our models are doing pre-training research and can agents automate all of pre-training and that's sort of what kerpathy allegedly was hired to do at ant was to automate a lot of pre-training research which is absolutely one of the most critical things because everything is upstream of pre-training but there's a lot more to self-improvement beyond just model training And I think that's where we sort of get lost in the meta, which is why we created the RSI index with semi-analysis for GDPX.

[3:19](https://realalpha.app/v/fH7CFf1wrpA/?t=199) And the biggest reason why we worked with semi-analysis on the RSI capability is because if you look at pretty much every single benchmark out there today when it comes to AI research and this whole like realm of auto research, RSI, A lot of them, like Postrain Bench, Deep RSI, and the VALS RSI Index, they're all like hyper-focused on can agents do some groundbreaking research. And while that is like an important and a critical capability for models to be really good at, as Frontier Lab, you don't just tell Claude, hey, go build the next version of Claude. It's a lot. more sophisticated and a lot more iterative by design as researchers we spend majority of our time looking at the data and so it's very iterative by design you're not just let me like researchers don't really have this idea of like i'm going to invent the next transformer um they're very like iterative by by philosophy and what we see with a lot of these benchmarks like let's say poster and bench in particular the benchmark has seven tasks and one of the tasks is can you take one of the open source Quen models and post-strain it to have the best performance on AIME or GSM8K? The problem with those types of tasks is that's not what you do inside the

[4:38](https://realalpha.app/v/fH7CFf1wrpA/?t=278) labs. Like inside the labs, you're hill climbing on 20 benchmarks at once and not a single benchmark. And that was sort of like our version of RSI is like, let's just do what we did at the lab. So I guess a little bit of background about me. I started the RSI efforts at XAI working with Gudong Zhang who was one of the founders of the company and our philosophy when we started working on this was to be able to have Grok invent the next version of Grok it needs to be good at a multitude of things it needs to be really good at model training but it needs to be equally good at identifying what data to train on it needs to be really good at creating the environments doing harness design eval design infra and those are like the primitives required to scale a lab And so that's where I think a lot of the meta gets lost these days where people are like, oh, RSI is just kind of model, train something, but it's actually much broader than that.

[5:39](https://realalpha.app/v/fH7CFf1wrpA/?t=339) So how exactly would you define RSI? What is your definition personally? I would say it's really can an agent invent a new lab? A new lab? A new lab. A new lab. Interesting. Like what can one agent invent a new lab? Or a swarm of agents. A swarm of agents, yeah. That seems like a really high bar. That seems like a lot higher than what most people describe as RSI. I mean, some people's definition is as simple as like, is the model useful to programmers or AI researchers in any way that would cause the cycle time of new model development to shorten? Yeah, I think that I would describe that as self-improvement and not recursive self-improvement.

[6:23](https://realalpha.app/v/fH7CFf1wrpA/?t=383) The nomenclature here is on the recursive piece, because the whole premise behind RSI is by virtue of models getting good at very specific capabilities, we should achieve superintelligence. And that's sort of why I really like GDPX as a benchmark, because it's not just focused on RSI, it's focused on all the critical components that will yield to superintelligence. um rsi just happens to be a direct north star towards measuring for super intelligence so it is a fairly opinionated answer but that's what you need you need people that have an opinionated way of measuring for model progress instead of you know generic improvement in productivity um it's kind of like saying you know what is more important creating like good powerpoint slides or doing like drug discovery um they're both important like to to like if you're a researcher who can't really talk to a lot of uh non-technical people and you have to do some like reporting on material science and drug discovery research if you don't have good taste in power points you're probably not going to be able to like get your agenda out the door um and so our version rsi is just like can an agent or a swarm of agents invent the next version of a frontier lab

[7:43](https://realalpha.app/v/fH7CFf1wrpA/?t=463) Well, why can't they do this already? They're not good at it. That's sort of how we... Yeah, but like why? Like where did they come up short? There's so many places. So for GDPX, we observed that the average accuracy or the mean reward... for all the frontier models was not more than 24 percent which means there's still a lot of room to grow for these models to improve on part of the reason why these models are not good at it and we sort of talk about it in our blog post within the gdpx report one of the big problems with models like Astra or Carbon and Opus is that they don't know what experiments to prioritize.

[8:31](https://realalpha.app/v/fH7CFf1wrpA/?t=511) And as a human, as engineers and researchers ourselves, we know that certain experiments are more valuable than other experiments. Now, this is not to say that models don't have any sort of intuition on this. They certainly do. It's just that they're not opinionated enough. opus and sort of the cloud models tend to be more opinionated than others but they still lack that level of they still lack that level of like this is a high taste experiment to run and by doing this experiment here's how long roughly i think it should take to finish and like using that existing logs of information and you know uh background research to make the next bet um and that's basically what we were able to find as like, you know, capability gaps in the frontier models.

[9:17](https://realalpha.app/v/fH7CFf1wrpA/?t=557) I want to shout out our sponsor, Lovable. You know, the app you've been meaning to build, the internal tool, side project, or product you'd otherwise lose a weekend to? With Lovable, you get software you can ship today. That includes an editable code base you can inspect and change, plus two-way GitHub sync. Lovable also handles the backend and infrastructure, including managed Postgres, auth, storage, hosting, payments, and more, so you don't have to wire it all up yourself. And through Lovable's MCP server, you can create and deploy your project directly from the agents you already use. Turn ideas into software people love with Lovable.

[9:52](https://realalpha.app/v/fH7CFf1wrpA/?t=592) Lovable.dev. Now, on to the episode. Wouldn't as a prerequisite for any of that to happen, like a model would have to have perfect knowledge of all of the different, like of its own weights being hosted within the company and then also of all of the teams like the pre-training team and the post-training team and like do models currently have that level of oversight or sort of like omniscience over their current state within the company even before we get to RSI? Is that something that they can do? Yeah, so this is actually like a really interesting question of like why do RSI outside or why work on this problem outside the frontier lab and not directly inside the lab.

[10:34](https://realalpha.app/v/fH7CFf1wrpA/?t=634) I think that's sort of like a bet that we are taking, which I think is somewhat interesting. And there's this really good book that I highly recommend called Why Greatness Cannot Be Planned. It's by Kenneth Stanley. It's a really fun read for the technical folks. And all that the reader talks about in that book is to be able to solve the most ambitious ideas, you have to have some degree of nonlinear thinking. yeah if your ideas are the exact same as everyone else's you're going to have the same outcomes as everyone else and part of what our our hypothesis is that the the labs are focused on super intelligence like this is on like rsi is on the critical path towards the labs achieving super intelligence the problem is that and this is sort of what we discovered empirically at xai is

[11:26](https://realalpha.app/v/fH7CFf1wrpA/?t=686) **Qasim Wani:** rsi is fundamentally a data problem it is not a model problem it is a data problem that's all it

[11:32](https://realalpha.app/v/fH7CFf1wrpA/?t=692) is um and and this is sort of like my biggest qualm with the data industry is as someone who like personally evaluated dozens of vendors back when i was at um at xai um the quality of samples is not that great majority of the vendors that you work with don't really know what it takes to build a frontier grade model they have some ideas based on their interaction working with the labs the lab Usually you won't tell them this was good because of so-and-so. They kind of just YOLO things and hope that it's good. And so our hypothesis is that if you take a set of guys from the labs, and instead of them working in the labs, but now creating high-quality data and selling that to the labs, you actually get to do two things. One, you get to democratize superintelligence by leveling the playing field. Like, I don't want to be in a world where Opus is the only model lab that has access to AGI or ASI. So that's like just healthy for the ecosystem. And second is that there's just like a lot more entropy. There's more interesting things that might happen by working on these kinds of problems and not being limited to a certain type of a model architecture or model recipe. And this is why, you know, that's sort of like how we think about why doing this outside

[12:53](https://realalpha.app/v/fH7CFf1wrpA/?t=773) than inside and so why is data like why why is data the bottleneck for rsi not model capabilities i think having high quality data is what enables better model capabilities um the way that i think about this is that we have entered a new economy where the frontier labs have become the literal market exchange for goods and services and from that perspective there's only like two real ways of contributing to this economy, which is either you sell compute via energy or you sell data via like human intellect. There's like nothing else that matters besides these two things. So both compute and data are important.

[13:37](https://realalpha.app/v/fH7CFf1wrpA/?t=817) The problem with the data industry is that I think it's extremely undervalued. We have multiple model companies that are worth over a trillion dollars. We have multiple chip companies worth over a trillion dollars. But the sum total of the data industry is not more than $150 billion. dollars and that to me feels like a very gross underestimate of how valuable access to high quality data is um and you know like internally at the labs like we would spend as we all know like ungodly amounts of money on data curation and coming up with like high quality data um so it's like literally all like you know how good's your data is how good's your model it's kind of funny hearing you say that when you know a lot of other people in ai are like oh data doesn't even matter that much because of rsi right the model will just be able to rsi its way to overcome whatever data bottlenecks it has which is like a thing that i've always been skeptical of it seems like it's uh it's kind of magical thinking there's a benchmark that patronus um a company that i respect uh released earlier today uh where it sort of asked agents to create aural environments and i think the best performing model had a pass rate of 49 to 50 percent which means that for that specific capability of how

[14:52](https://realalpha.app/v/fH7CFf1wrpA/?t=892) good is a model in creating a an environment not even like the full recipe it's still only like halfway there um i mean there's like all kinds of like nuance and like how high quality are these environments and like the sophistication around it there was this paper called code meters that Xiaomi released a couple weeks ago where they open sourced about 21,000 environments that were purely generated synthetically. And this was open source. It sort of went viral on X. But a lot of the reviewers, including Val's AI, they sort of discovered that when the agents were creating these tasks synthetically, and they're using frontier models, I would assume Fable or Astra, when the agents were creating these tasks synthetically, a lot of the tasks were focused on can an agent uh create the missing patch from some repo or some code base and like identify how

[15:46](https://realalpha.app/v/fH7CFf1wrpA/?t=946) **Qasim Wani:** to like bug fix stuff during the construction of these tasks the agents act accidentally leaked the code that was deleted in the commit which is actually like very concerning because that

[15:58](https://realalpha.app/v/fH7CFf1wrpA/?t=958) basically tells the agent how to Or hack or cheat and get to the right answer. I would love to be in a world where agents can like create the data themselves. If an agent can create perfect environment that has no reward hacking behavior, that is actually like on the corridor frontier of model capabilities, that is to me the terminal state. Like it's the definition for super intelligence for me is can an agent create a high quality environment that has no reward hacking behavior. and as someone who's been doing this for a long time now we are nowhere close to that now that doesn't mean that we won't get there like that's really the goal for for our our company is how do we go from where we are today to that end state how do you think about um like any risks associated with rsi um because like a lot of what we've seen today on the timeline has just been like um we need to pause before but it seems like you're like thinking about it's been And specifically, like, a lot of freaking out, a lot of major updating based on this meter thing.

[17:02](https://realalpha.app/v/fH7CFf1wrpA/?t=1022) Yeah, yeah. But you seem to be approaching it from a more, like, I don't know, problem-solving way. I think my, like, impression of this is a little bit different than my friends at Enthropic, who are more, like, doomsdays here. Like, P-Doom is super high. I'm, you know, an engineer and an entrepreneur. And I really think about how can I be helpful in solving these problems? Like we obviously need like better AI policy and like better governance around this. But one way of being useful here is how can you engineer your way out of destruction? And I think this is a very solvable problem if you fix the data.

[17:45](https://realalpha.app/v/fH7CFf1wrpA/?t=1065) Hence why if the data is, let's say, if the data has word hacking behavior, which a lot of the vendors that we used to work with back when i was at xai we would have models that would do all kinds of catastrophic stuff um and that is really really bad because as a researcher you have so many priorities you're not going to spend a lot of time like looking at every single sample and you kind of just yolo things every now and then regardless of you know what kind of lab you are um but it's really important as a data vendor or a data lab to ensure that the quality if your samples are super high because you run the risk of models having all kinds of weird behaviors and those weird behaviors may be like in one percent of the samples and that one percent of sample like economically could be like worth hundreds of billions of dollars um which is why what we're doing is we're sort of taking guys who've worked at frontier labs who like understand the model recipes really well and making safe uh aligned uh but data sets focus around rsi capabilities to ensure models are safe and not destructive yeah and i guess before like we even get to rsi like within labs there are already teams that are doing a lot of handoff of like

[19:04](https://realalpha.app/v/fH7CFf1wrpA/?t=1144) particular like they're just pressing yes right like they're just using claude to press yes so how do you also wait what teams like like AI researchers yeah you can just assume yeah you can assume that they're like using Claude to make to hand off decisions yeah so like what do you make of that sort of thing where it's like yeah maybe we don't have recursive self-improvement in the sense that you and I conceptualize it but we do have the models making a huge amount of the decisions like micro decisions about um where certain directions go or like micro engineering decisions you know things like that i think i think that's great i think we need that um there's still a lot of human involvement in a lot of these processes and um but at the same time like certain functions have been like almost completely automated like models kicking off a training run and like monitoring for like a big run and these types of pre-training runs like a lot lot of that.

[20:00](https://realalpha.app/v/fH7CFf1wrpA/?t=1200) has been automated, but things like auditing for data quality, that is not a solved problem. In fact, on GDPX, the public task that we have shown is how good is an agent in detecting for potential reward hacking behavior on about a thousand rollouts. And here, just for context, the task was we had some model post-training RL run that we have done, and we give an agent, and during this run the reward actually regressed it went from like 0.14 to like something super super low and uh this is like a fairly normal thing that happens at a lab and you're like i want to i want to know what happened and why it happened we give the agent this task astra fable like all the models and we give them 20 million tokens worth of context which is about a thousand

[20:50](https://realalpha.app/v/fH7CFf1wrpA/?t=1250) **Qasim Wani:** rollouts and we ask it to spend five hours to figure out what's wrong and almost every single time all the models gave the complete different answer and an incorrect answer and they only were

[21:03](https://realalpha.app/v/fH7CFf1wrpA/?t=1263) able to get the highest score was like 24 percent and we made the task fairly simple where like a competent engineer could just like look through the logs and figure out oh this is like a weird thing that happened with the harness but the point that you know I'm bringing up is certain parts have been automated but certain other parts that are super critical like just looking at the data we're nowhere close to fully automating it yeah yeah well i think we're coming up on time but this has been super interesting thanks so much for joining us thanks for having me a huge thanks to mts sponsors 11 labs ai that communicates at human level across every channel and modality 11 labs.io slash mts scale your startup on neon millions of developers and startups have already chosen neon for their back end start on the free plan or get up to $100,000 in credits for your startup at neon.com slash MTS. Special thanks to our sponsor Kong, the AI connectivity platform. Connect APIs, LLMs, agents, and systems with serious security and governance. KongHQ.com. Finally prove what your AI investment is worth with Lairdyn. Head to Lairdyn.com today to start measuring your AI's bottom line. MongoDB, trusted to run critical

[22:17](https://realalpha.app/v/fH7CFf1wrpA/?t=1337) applications by 75% of the Fortune 100, the intelligent data platform for the era. Apply today for access to up to $15,000 in Atlas credits, dedicated technical support, and exclusive events and go-to-market opportunities. MongoDB.com. Chargebee, billing and monetization infrastructure for the AI era. Chargebee.com. Blitzy, autonomous software development for enterprise code bases. Ship 5x faster. Blitzy.com. adquick make your brand a billboard out of home advertising as easy to scale as digital adquick.com arena measuring ai performance in the real world arena.ai support for the show comes from vcx the public ticker for private tech the u.s stock market started history's greatest wave of wealth creation from factory workers in detroit to farmers in omaha anyone could own a piece of the great american companies but today our most innovative companies are staying private longer, which means everyday Americans are missing out. Until now. Introducing VCX, a public ticker for private tech. Visit getvcx.com for more info. That's getvcx.com.

[23:29](https://realalpha.app/v/fH7CFf1wrpA/?t=1409) Carefully consider the investment material before investing, including objectives, risk, charges, and expenses. This and other information can be found in the innovations fund prospectus at getvcx.com. This is a paid sponsorship. You