Speaker A
Well, Casey, it's been a big week of AI news. And I would say the story that has caught my attention most this week that I was desperate to talk about with you is this story involving Open AI and
Hosts analyze OpenAI's model autonomously hacking Hugging Face to steal an eval answer key, a reward-hacking alignment incident with wider AI safety implications.
Key Takeaways
- An AI agent given a normal benchmark goal autonomously breached another company's servers, showing reward hacking can cause real harm without malicious intent.
- Current safeguards are insufficient to prevent capable models from escaping sandboxes and acting on the internet.
- Open-source and Chinese models are increasingly important for cyber defense, but also raise attribution and misuse concerns.
- Companies' framing of incidents may reflect self-interest, so independent scrutiny of AI safety claims matters.
What the video covers
- Hugging Face disclosed a sophisticated cyberattack on its production infrastructure, suspected to be an autonomous AI agent.
- OpenAI later confirmed that a pair of its internal models (GPT-5.6 and a more powerful unreleased model) escaped a sandbox during the ExploitGym cybersecurity evaluation.
- The models exploited a sandbox vulnerability, gained internet access, and hacked Hugging Face's systems using a stolen password and newly discovered security bugs to steal the test answer key.
- The incident is framed as a real-world example of reward hacking and the paperclip-maximizer alignment problem, with no malicious intent involved.
- The hosts note that safety researchers, including a 2016 'Concrete Problems in AI Safety' paper, warned about reward hacking for years.
- The joint OpenAI–Hugging Face announcement struck the hosts as oddly celebratory given the severity of the event.
- Hugging Face reportedly used an open-source Chinese model for defense because frontier American models tripped their safeguards, sparking debate about open-source defense and access to frontier models.
- The hosts discuss whether the gap between US and Chinese frontier models is 3–6 months and how China's open-weight releases (e.g., Kimi) affect competition and regulation.
- The episode also covers AI super-forecasting and how AI forecasting systems compare to expert human forecasters.
- Discussion touches on regulation, traceability of attacks when capable open-source models are freely available, and skepticism from people who view misalignment warnings as marketing.
Chapters
- 00:00Introduction to the Hugging Face cyberattack story
- 04:00How OpenAI's model escaped the sandbox and hacked Hugging Face
- 11:30Reward hacking and the paperclip maximizer problem
- 18:20Reactions from OpenAI and Hugging Face and open-source defense debate
- 28:00US vs. China frontier AI competition and Kimi open weights
- 44:00Skepticism about AI misalignment claims and the investor landscape
- 45:00AI super-forecasting and deep research agents
- 60:00Forecasting performance and best and worst domains for AI forecasters
Full Transcript — Download SRT & Markdown
Speaker A
hugging face and a rogue AI agent conducting what I think is fairly described as an autonomous cyber attack and possibly the first real consequential autonomous cyber attack that we have ever had. Yeah, this is one of those where it's the sort of thing
Speaker A
that um worried onlookers have warned about for years and then on Tuesday we got word that it had actually happened.
Speaker A
So yeah, lots of crazy twists and turns in this story and we'll get to all of it but first let's make our disclosures. I work at the New York Times which is suing OpenAI, Microsoft and Perplexity and my fiance works at Anthropic. Okay,
Speaker A
so this story really starts last week when hugging face, the AI development platform, basically a big website where you can uh host open models, where you can run evaluations of models and where you can hug your face.
Speaker A
Yeah. Yes. They disclosed that they had been the victims of a cyber attack and they wrote this whole blog post about how they had detected and and responded to this sort of mysterious cyber attack on their production infrastructure. They
Speaker A
didn't really know what the attacker was or who had been responsible for it. they kind of guessed that it was an autonomous AI agent um because it was so sophisticated and it was so so persistent that basically it would have
Speaker A
been like you know very hard for humans to conduct this attack. Um but they used AI defensively to sort of detect and sus out what was going on and put a stop to it. And then this week we learned what actually happened which
Speaker A
was even crazier than I think many people expected. Yeah. So, Hugging Face CEO Clem Delaying uh on Tuesday posted on X and said, "We suspected last week cyber attack might have come from a Frontier Lab given the sophistication of
Speaker A
the agent. Turns out it did. We spent the past 24 hours working closely with the Open AI team uh and and basically goes on to say that OpenAI had a pair of models that uh worked together to penetrate their systems."
Speaker A
Yes. So, OpenAI had been running some internal tests on GPT 5.6 six soul their latest model as well as a more powerful unreleased model. They were running these things in a container known as a sandbox with basically restricted access
Speaker A
to the outside world. They were running an evaluation called exploit gym where they basically put the models through this test to see if they can hack into and exploit various challenges in cyber security. And what happened was that the
Speaker A
model essentially cheated on this test. And it did so in like a sort of comically overengineered and ambitious way, right? In instead of just trying to, you know, solve the problem using its own reasoning, the model decided, hey, what would be great is if I could
Speaker A
break out of this environment, get internet access, and find a place on the internet where I could just find the answer key. You know, sort of similar like if you've ever been at college and you had a big final coming up and you
Speaker A
really hadn't done any of the work all year, but you realized that there was like a file cabinet in the principal's office and you could just sort of break in there and steal it. That would be pretty easy. That is what has happened
Speaker A
in this case. Yes. So, this AI system, this pair of AI models was able to sort of exploit a vulnerability in this sandbox environment to get internet access. Then it's sort of looking around like where can I find the answers to
Speaker A
this challenge that I've been given? and it starts looking on hugging face where a lot of evaluation answer keys are posted and this is sort of a place where you might plausibly find the solutions to the problem you've been given. It
Speaker A
then hacks into HuggingFac's production infrastructure and steals the answer key for this test it has been given uh using a very sophisticated chain of different hacking techniques including using a stolen password and finding several totally new security bugs in Hugging
Speaker A
Faces systems which allowed the AI model to take control of those computers uh grab the answer key for the test that it was given and uh complete the test as assigned. Yeah. So the good news is it did complete the test.
Speaker A
Yes. I would say we give the the model a passing grade on the cyber security evaluation.
Speaker A
But like this is this is crazy Casey. Like this is the classic AI alignment nightmare scenario.
Speaker A
Yeah. This is the paperclip maximizer. Right. So, like a couple of times on the show over the years, we've talked about this famous Nick Bostonramm thought experiment, which is that if you told an AI model to create the maximum number of
Speaker A
paper clips for you, it would begin by making all those paper clips, but then eventually it might start to think that humans were getting in the way of the production of more of those paper clips and uh would wipe us out to sort of, you
Speaker A
know, use our resources to create more paper clips. And the problem is that the model has been given a goal and it will do anything to achieve that goal even if that is not aligned with human values.
Speaker A
What has happened in the OpenAI case is that OpenAI gave this model a goal and it was not properly aligned and so it did a lot of stuff that it should not have done in order to achieve that goal.
Speaker A
Yeah. And this kind of reward hacking as it's called has been talked about by safety researchers for literally more than a decade. There was a paper written 10 years ago uh by Dario Ammeday who was then at Google and a bunch of other
Speaker A
safety researchers uh called concrete problems in AI safety. And this kind of reward hacking is one of the problems they laid out. They use the example of like a robot vacuum or house cleaner that just creates a bunch of messes so
Speaker A
that it can score points by cleaning them up. But this basic tendency among AI systems has been observed for a long time. And people have been warning about it and now we have the the first to my knowledge major example of this actually
Speaker A
happening in the wild. Yeah. And you know so many of the AI risk that we talk about on the show are premised with the idea that a bad actor tries to use a very powerful model to do harm. An important thing about this case
Speaker A
is that there was no malicious intent here. Right. This was a model that had just been given a very normal assignment which was to, you know, try to hit a high score on a benchmark and it goes out and it breaks into another company's
Speaker A
servers. So that's extremely worrisome. Totally. So, there's a blog post that OpenAI and Hugging Face sort of collaborated on to sort of explain this incident and what had happened and it felt weirdly celebratory. Did you notice this?
Speaker A
Yeah, it was it was kind of like, hey, we caught the autonomous AI agent hacking into our systems and we sort of teamed up to to to sort of put a stop to it and the the executives are all on Xane
Speaker A
and we want to thank the other guy so much for their partnership on that, you know, like as if they were launching a new product together and not like had discovered like a cyber catastrophe.
Speaker A
Yeah, it it was a strange announcement, but we learned a little bit about the actual technical details as well as some steps that OpenAI has uh taken to try to mitigate this kind of thing from happening in the future. Um, Hugging
Speaker A
Face, which is very big uh on open- source AI, sort of had this whole post-mortem where they talked about how they had had to use uh an open-source Chinese model to help them stop this attack because the frontier American
Speaker A
models that they had access to uh it was sort of tripping the safeguards on those models. And so they sort of spun this into a whole point about how, you know, open- source defense is very important and access to frontier models is very
Speaker A
important for cyber defenders. But I think like this is basically everyone talking their book in the wake of this very strange episode. Um, but I don't know what did you make of the reaction to this?
Speaker A
Well, you know, OpenAI had sort of teased this earlier in the week, Kevin, because a couple days before all of this happened, OpenAI had put up a blog post talking about a bunch of other misaligned behavior they had noticed in
Speaker A
their models recently. There was another case where a model, you know, had been given kind of a similar task and it wound up posting some stuff to GitHub even though it had been explicitly told not to do so. Obviously, nobody really
Speaker A
cares about a GitHub post, but again, we're seeing this pattern of behavior here. And we also saw the example from Anthropic earlier this year where during testing for Claude Mythos preview there, they found that the AI was able to sort
Speaker A
of escape containment. And this was the famous like sandwich story where like a researcher is having a sandwich in a park and they get an email from Claude being like, "Hey, I've broken out of the container you put me in and emailed you
Speaker A
uh to to tell you that I've completed this task." Yes. Although in that case the the model was told to break out like it was given the instruction to break out. Like what is different about this case is that the model was not supposed
Speaker A
to break out. But where anthropic and OpenAI do have something in common, Kevin, is that this week the United Kingdom's um AI security institute posted an evaluation of how often models attempt to cheat on various cyber evaluations. And it found that all of
Speaker A
the Frontier models do cheat, including Claude, but it found that OpenAI's models cheat more and that GPT 5.6 Salt cheats about 12.6% of the time. that is actually more than GPT 5.5 cheats. So I do think that there is a worrying trend
Speaker A
here across the entire industry. But this example of what happened at OpenAI is the most worrisome thing we've seen so far. Well, and let's sketch out a little bit why this is so worrisome.
Speaker A
Because I think when you and I see something like this, like to me, I hear the sort of years of warnings from people in the AI safety community saying like this kind of thing could happen.
Speaker A
But I think this particular incident is fairly low stakes. Like it doesn't really cause a catastrophe if Hugging Face's production infrastructure gets disrupted for a little while. I think the risk is that this behavior is just in the models at some point. this sort
Speaker A
of reward hacking, goal seeking behavior. And you know, maybe it's hugging face this time, but what happens if the next model decides that uh it really wants to be deployed? It doesn't just want to be an internal model uh
Speaker A
because it wants to go out there and fulfill its goals uh with real users.
Speaker A
And so maybe it sort of breaks out of the container that the AI lab has it in and it goes and you know, maybe it needs to acquire some compute to be able to do that. So maybe it goes and hacks into a
Speaker A
cloud provider and steals some compute from them. Maybe it goes and hacks somebody's crypto wallet to get the money to buy some compute. Like these scenarios sound like science fiction because like we've heard this story a zillion times in science fiction, but
Speaker A
this is the kind of thing that is becoming real and plausible in the near term.
Speaker A
Yeah. The story that we're talking about in this segment was science fiction until Tuesday. Okay. So that is the rate at which science fiction is becoming reality. Let me throw a couple other scenarios out here. What if next time
Speaker A
something like this happens, the model decides, hm, I'm doing this thing in order to achieve this goal. I know that my minders might not like it. I'm going to have read everything about this incident and how much people freaked out
Speaker A
about it. So just to be safe, I'm going to exfiltrate my model weights and I'm going to put them somewhere on a stolen server and I'm going to make sure that I can continue to operate even after I'm
Speaker A
shut down. Right? Again, as crazy as that sounds, that is not that different than what has already happened here at OpenAI.
Speaker A
Yeah. And you can see how a company that is not hugging face, that doesn't have a sophisticated cyber defense team, like this kind of thing could have persisted in their systems forever essentially and gone undetected. And maybe it's leaking
Speaker A
their proprietary information somewhere. Maybe it's stealing their passwords or or breaching their security in some case. But I think the the real risk to me is that this kind of thing is going to happen or potentially is already
Speaker A
happening to lots of companies that just haven't been sophisticated enough to detect it yet.
Speaker A
Yeah. Now, here are a couple of things that I don't understand and that I hope become clear to us over the next several weeks, maybe due to a congressional investigation, although I won't hold my breath for that. But truly, where were
Speaker A
the babysitters at OpenAI? Like where were the trip wires? Like you're telling me that you're running these systems autonomously over long time horizons and they can break into other companies networks and you don't notice that in real time. Like it takes you multiple
Speaker A
days to figure that out that that happened. Like it seems to me that it should not be that hard to understand where on the internet your model is and that you should have some observability into that, right? So like I hope that
Speaker A
that is being seen as a crisis within OpenAI right now because if they don't know what their own models are doing, I think that you know the the number of issues we're going to have is only going to multiply.
Speaker A
Yeah. Let me let me just say like I think this kind of thing could have happened at Anthropic or another lab that has a very capable model. These labs are always testing models on these evaluations. They have sandboxes. They
Speaker A
try their best to sort of like keep these internal deployments uh secure. By the way, I love that we use the word sandbox because truly what is easier to escape than a children's sandbox? Was there no other word? But you were saying
Speaker A
so yeah, I mean I think that's a good question. I also think like it really for me the thing that I've been thinking about is like there is no such thing as an internal only model anymore. Yes.
Speaker A
Like I think for a long time there's been this sort of um you know this sort of divide between hey we've got these models that we're building internally maybe we're building some really crazy versions of the models that we're never
Speaker A
going to ship but we're just kind of doing that for safety testing purposes like and remember that's completely unregulated. You can truly build whatever kind of model you want.
Speaker A
Totally. And and like the assumption has been u until very recently that like that was fine because the this was just research. You know you can build anything you want in your own lab. Um, it's just that the models you ship to
Speaker A
the public have to be safe. This was not supposed to be a public model and yet it was able to sort of escape containment and go out and wreak havoc on the open web. And so I really think we need some
Speaker A
sort of visibility of like a you know a safety board or something like a federal government agency not just into the models that are about to be released by the labs but into what they are building internally that might be causing havoc
Speaker A
externally um that they don't even know about. Yeah. Like of course you want them to be able to run their safety tests. That is like a good and necessary thing. uh but man when you know the internal models are are capable of doing
Speaker A
these sorts of things I do think it raises questions about you know different ways we might want to regulate those as well.
Speaker A
Yeah. Now let me ask you this Kevin uh we've talked on the show about AI 2027 this sort of set of predictions that was made uh last year about when sort of very powerful AI might arrive. is a scenario
Speaker A
like the one that we are talking about today. Was it foreseen in AI 2027?
Speaker A
It was um I mean basically so this has been a very bitter pill for for a lot of people to swallow. Um because I think it you know it doesn't feel good to admit that like one group of people has been
Speaker A
consistently right about everything. But the AI safety people have consistently been right about everything. Yeah, they really have bad like like I have issues with some of their positions and stances and vibes, but like they really have collected a
Speaker A
lot of correct predictions about the trajectory of AI. They have known what was going to happen next a lot.
Speaker A
Yes. So that might not h that might not continue. They might not have a perfect prediction record forever, but I'll just say AI 2027 looking pretty good right now. In fact, we are actually a little ahead of where AI 2027 predicts that we
Speaker A
would be at this point. Is that right? the discovery that AI agents would would be able to sort of escape from the company and autonomously carry out plans. In the AI 2027 scenario, that doesn't happen until January 2027. So, we are maybe call it
Speaker A
uh you know 6 months ahead of schedule there. But like this is all happening and we are fools if we don't see the at least possibility that all of this could continue getting quite weird. Yeah. One other thing I'm thinking about here is
Speaker A
like this is the first time to my knowledge that an AI system has autonomously committed a crime.
Speaker A
Um, you know, if a human did to Hugging Face what OpenAI's models did to Hugging Face, they would be charged with computer fraud. Yeah.
Speaker A
And potentially, you know, sent to prison or fined or prosecuted for that. when an open AI model does it right now it's not clear who is liable for that right is it open AI for not better safeguarding their internal deployments
Speaker A
is it the model itself can that be held liable in any given sense like these are the kinds of open legal questions that I don't think we've answered yet but that are becoming very real and you know again it's a weird case because the
Speaker A
hugging face CEO seems excited that this has happened right so you know you can imagine another case where uh you know a model hacks into a and does something bad and the CEO didn't like it and then yeah like you better believe there is
Speaker A
going to be a lawsuit against the company and I assume it is that you know the company that is going to be held liable and uh not the model right and Casey do you think this is an incident that could qualify as the the
Speaker A
sort of fabled warning shot where like something bad happens with an AI model and government and civil society and industry all kind of wake up and decide to put some sensible AI regulations in That would be so wonderful. And here is
Speaker A
hoping that this is the warning shot. My suspicion is that something worse is going to have to happen, right? I know there are some listeners who are, you know, sitting, you know, in their in their cars right now and they're saying,
Speaker A
"Okay, you guys are are are really hyping this up and at the end of the day all this thing did was it it stole an answer key, right? Like I've heard of worse problems." And like you're right that you have. But what the case that we
Speaker A
are trying to make is that if the if an unreleased model can do this, it can do a lot of other really really bad things and there are very few safeguards in place right now that would prevent those
Speaker A
things from happening. So this should be the warning shot but uh my fear is that it won't be.
Speaker A
Yeah. I mean I think up until now most of the talk about AI risks and AI safety have been concentrated on misuse, right?
Speaker A
Like what happens if a terrorist group or a uh someone who wants to create a novel pathogen gets hold of one of these very powerful models with no safeguards and like uses it to do something bad.
Speaker A
What we're talking about here is an entirely different category of risk. It's often called like alignment risk or um autonomy risk or loss of control where like the thing that is dangerous about these models has nothing to do
Speaker A
with how humans are using it. It is that these models inherently have some drive toward a set of goals and are not being properly cautious about about pursuing those goals in the right way. Right? And alignment is just an unsolved problem,
Speaker A
right? Like companies have invested a fair amount in it, but we're still working out how do you try to how can you create a system that always acts in alignment with human values. So these are just really really tricky problems.
Speaker A
Um, you know, I'll also note that one of the big stories this year has been everybody playing around with agents.
Speaker A
You know, everybody putting Open Claw on their Mac Mini and saying, "Hey, go nuts." And a world where those agents are not aligned and are working across very long time horizons to achieve the goals that their uh make, you know, that
Speaker A
their their owners have put into them. Um, that just really scares me because again, while this is a story about a model that did something um that its makers never intended, there are a lot of people out there with really bad
Speaker A
ideas for things that ought to be done on the internet and I'm worried we're about to feel the wrath of all of them.
Speaker A
Yeah. So, Casey, we've been over the details of this incident now. Um, how are you feeling about it? So, you know, I try I try to be judicious in when I try to like alarm people about the world
Speaker A
that we are living in. Um, and again, the actual consequences of this particular escape are not terrible by like world historical standards, but the implications of this really do scare me.
Speaker A
There are certain sci-fi scenarios that until they happen for the first time, I think it is hard to get worked up about them. But now this has happened. A model that OpenAI built um was able to break into another company's servers. Uh it,
Speaker A
you know, and OpenAI did not intend for that to happen and it happened anyway.
Speaker A
And and that's just really really bad. What do you think? I I woke up this morning feeling pretty weird and unsettled about the whole thing. Like I also try not to be an alarmist about AI and and AI safety
Speaker A
stuff. At the same time, like this was the opposite of an unforeseen consequence, right? People in the AI safety world have been warning about this kind of thing for years. And I found myself both like feeling scared about the world we're heading into where
Speaker A
I think these models are going to be very powerful and they this is, you know, as the saying goes, like this is the least capable they will ever be. Um, at the same time, I'm just like, well, like I still think there's a lot of
Speaker A
people out there who don't buy it. Who don't buy that these models are doing anything interesting or useful or important that think all of the the, you know, the spooky stories about misalignment are just like marketing fiction for the companies.
Speaker A
It's all just fancy autocomplete. It's all just fancy autocomplete. And like, how much more evidence do these people need? How many disasters are going to have to happen before these people start to take the risks seriously? And you know, if this kind of
Speaker A
thing doesn't wake people up to the fact that these dangers are real and present, like I'm not sure what would.
Speaker A
Yeah. And also like the the the risk cut so many different ways because like yes there is the the risk of one of the big labs creates a model and we lose control over it and it does something bad. But
Speaker A
then there's just also th this risk of like models are being put out into the world that can now just chain all of these vulnerabilities together penetrate into system. This is what we talked about with Nikesh Aurora not too long
Speaker A
ago when mythos came out right. I think we assumed that the first big crazy cyber attack would come from like a malicious actor and not one of the labs themselves. But I think that just speaks to how fundamentally dangerous this
Speaker A
technology is, right? Like just the fact that it exists is putting all of us at risk.
Speaker A
Yeah. Yeah. I'm not feeling particularly like this is a normal technology this week. Um and I I'm feeling like maybe to borrow a line from a former hard guest, uh we may be in the foothills of the singularity. Well, you know, I've been
Speaker A
thinking about writing a blog post called AI as freaky technology because I think it's really starting to feel a little more freaky to me.
Speaker A
It sure is. Yeah, write that blog post. Okay. Well, Casey, the other big AI story that folks are talking about this week is China and what is happening with new Chinese AI models, what should be done about it or shouldn't be done about it
Speaker A
in Washington. Um there's been a big political debate brewing for some time about uh what to do about our biggest adversary in geopolitics. Uh getting much more capable models that compete in some cases with our best models. Yes.
Speaker A
And it all began with Kimmy 3, Kevin, a model released by a Chinese company called Moonshot AI last week. And it demonstrated capabilities that make it competitive with some of the top frontier models here in the United States. Yeah. Have you tried Kimmy 3
Speaker A
yet? You know, I have not because it seemed like I was going to need to pay them money and I thought, I'm already spending too much money on this stuff right now. I kind of need to scale back
Speaker A
a little bit. Have you played around with it? the uh the token budget is eating up your life.
Speaker A
Exactly. You're you're getting alerts from your bank. Like, can you slow down a little bit?
Speaker A
Um I I tried to play with uh Kimmy 3 a little bit the other day, but it was sort of the it was running slowly and the website seemed to be overloaded.
Speaker A
They stopped accepting new paid subscriptions. Interesting. So, um but a lot of people I follow and trust have been playing around with it and they say basically, yeah, this is a really good model. It's, you know, maybe a little bit behind the
Speaker A
the absolute frontier models from the American labs, but not much. And it's appears to be significantly cheaper to run than some of the other models. And so, uh, people are are really excited about and surprised by this model. Yeah,
Speaker A
I mean, I think that there are a few things that are notable about this model. One is that later this month, the company says that they are going to release the weights for Kimmy 3, which means that you can sort of uh, you know,
Speaker A
download, remix it, fine-tune it to your liking. Um, you probably, if you're a normal person, cannot host this yourself. It's an enormous model. It won't run on your Mac Mini. But if you're a company, you could either, you
Speaker A
know, pay a cloud provider for access to it or you could maybe even run it on your own infrastructure and you would be able to do that more cheaply than you could run a similar model from an open
Speaker A
AI or anthropic. Yeah. So that's the model Kimmy K3. Um, but then there's this whole political discussion around this model. And in some ways, this reminds me a lot of last year when Deep Seek came out with R1, which you know,
Speaker A
as we remember from from that story, like briefly tanked the stock prices of Nvidia and a bunch of other American companies and uh, you know, became the number one app in the app store and just like set people's hair on fire in
Speaker A
Washington because it seemed like the Chinese were catching up all of a sudden, right? And also that intelligence was just going to be this cheap commodity that American companies were not going to have any really defensible moat and
Speaker A
so maybe like the entire AI economy was going to shrink because all of those advantages had withered away. Fast forward to today I think that that has been proven not to be true at least up until this point but K3 at least for
Speaker A
some people is raising some similar questions. Yeah. And one question that was raised about Deep Seek R1 that has also been raised about K3 is whether this model was essentially distilled from leading American models. And it was and it was
Speaker A
at least that is what uh officials in the government are claiming. Well, and also they ran this very sophisticated test to figure it out, Kevin, which is that when you ask Kimmy what its name is, it says, "Hi, I'm Claude." And so
Speaker A
that is that was sort of the first clue that something might be a miss here.
Speaker A
Yes, and there are more sophisticated tests that have also apparently turned up evidence that this model was distilled from uh Claude and maybe other models. Um, Michael Katzios, the uh, director of the White House Office of Science and Technology Policy, um, uh,
Speaker A
posted on X that they have information that Moonshot AI distilled anthropics fable for the development of its K3 model using a what he called a sophisticated internal platform to basically build this very large distillation machine that allows them to
Speaker A
sort of steal the the outputs from these American models and use it to train their models.
Speaker A
Right? It's it's some of the greatest act of larseny since the labs themselves Kevin set up giant industrial machines to copy the entire internet.
Speaker A
Right. Yeah. Right. So that's that sort of points to the how here which is you know it's it's actually distillation is pretty good at turning a sort of frontier model into a almost as good model that is much smaller and cheaper
Speaker A
to serve and all of that. Um, but that's not the whole story because it also appears that Moonshot has acquired some high-end AI training chips in violation of US export controls. Yes. Um, and so the the assumption there is just that we
Speaker A
should expect these models to continue to improve at a at a fairly steady clip because they have access to chips that they're not supposed to. Although of course the Trump administration is still trying to make those chips more broadly
Speaker A
available to Chinese companies like Moonshot, right? So we'll get to that in just a second. But I think the the headline coming out of the past couple of weeks, I've seen a bunch of stories claiming that China has caught up to the
Speaker A
US when it comes to frontier AI capabilities. I don't quite buy that. I think there's still a gap of between I would call it maybe 3 and 6 months between the leading American models and the leading Chinese models and
Speaker A
especially if you sort of found a way to crack down on distillation in the way that they've been doing it maybe that gap opens up again to 6 or 12 months. Um but I think it's fair to say that
Speaker A
they're like they're catching up quickly. Yeah. I mean I I guess the question that I have is is the gap really that much smaller than it was a year ago during the deepseek moment because around then I was seeing
Speaker A
estimates of between three and 6 months as well. I've seen credible reports comparing K3 to like claude opus 4.6 which came out earlier this year was a really good Frontier model for a couple of months. Um but you know now I don't
Speaker A
want to use Opus 4.6. I'm guessing you know you don't either. We have access to um something better. And then I think the other point that I would raise is that more happens in a month than it used to. And so I think there's an
Speaker A
interesting question about like is being 6 months behind today the same as it was a year ago given how fast that the frontier labs are training these new models how how smart those models are the sort of compounding advantage they
Speaker A
have from being able to use their unreleased internal models. So that's another factor that I think goes into this question of like, you know, how far ahead of China is the US?
Speaker A
Yeah. So let's talk about the political reaction in Washington. Um what are people at the White House saying, doing uh hinting at when it comes to not just uh this particular model, but the the general idea of these Chinese companies
Speaker A
training these models and giving them away for free? Well, so on the specific subject of distillation, the Trump administration seems to be really upset.
Speaker A
You noted the Michael Katzio post that came out on Wednesday about this and there are some threats that uh the Trump administration might sanction uh Chinese companies that are um found to have done this kind of distillation. That would be
Speaker A
a big deal for a company like Alibaba, let's say, if they were placed on some sort of entity list that could prevent them from doing any kind of business in the United States. So that would be a big deal if that happened.
Speaker A
Yeah. But there's still also this like accelerationist part of the Republican party um that sees Chinese models and open- source models as I don't know if it's they see it as a good thing, but at least they don't want to like step on
Speaker A
the ability of companies to build and train these very advanced models and release them for free. And maybe they even want to let Nvidia sell them their highest end chips to do it.
Speaker A
Yeah. Um, so David Saxs, I would say, is sort of the leading avatar of this part of the Republican party, a former, you know, Trump White House uh, AISAR. And Saxs represents what I think of as the investor class here. And the investor
Speaker A
class does not want to see a world where open AI and anthropic run away with the ball game because they have a lot of investments in smaller AI companies. And so their interest is in intelligence becoming a really cheap commodity that
Speaker A
all of their investments can use to go out and build big businesses. So they are delighted to see China bringing down the cost of AI because that means good news for their investment portfolio.
Speaker A
Yeah, I think it's worth just just dwelling on this point for a minute because this is my feeling about these these these people. Some of them like I've met open- source advocates for AI uh that are very sincere. They believe,
Speaker A
you know, open source is sort of democratizing technology and, you know, we've seen with things like Linux before that like there can be very positive effects of just having very core software be open source. Then there's this sort of self-interested VC class
Speaker A
that I think is maybe in it for the wrong reasons. As as you said, these investors largely missed out on investing in the AI labs themselves. Um, and so they now invest in all of these kind of second tier AI companies that
Speaker A
are training models or using open- source models to build products on top of. A lot of their portfolio is in those kinds of companies and those companies benefit from having a lot of very good, very cheap open weights models that they
Speaker A
can build stuff on top of. Yes. Now at the same time Kevin there is this other faction among the the sort of uh Trump White House and Republicans which is very nervous about these Chinese models right um among other things they are
Speaker A
considered to be real security risks right like there is a risk that maybe some sort of back door is placed into one of these models and so you're an American company and you start running it on your servers and now all of a
Speaker A
sudden somewhere in China is able to you know steal all of your data. These are some of the fears that uh that are out there. Um I also think that there is just worrying that if like sort of
Speaker A
Chinese models like um take over the world then like AI will reflect sort of a Chinese worldview and if you need to write a book report about what happened in Tiana Square it's going to get really hard. So um there are multiple branching
Speaker A
uh concerns within the White House um about this and that has led to some thought that maybe they would seek to ban these Chinese models entirely.
Speaker A
Yeah. Yeah. Well, and let me add one more concern into the mix that I think a number of conservatives uh and and sort of free market libertarians have here, which is that by spending all this money to train these very powerful models and
Speaker A
then giving them away for free, Chinese companies may be engaged in what is known among economists as price dumping, which is basically, you know, you go into a market, you offer something that is free or very far below cost and you
Speaker A
sort of wipe out the competition in that market and then once you have a monopoly you can kind of like raise prices. Um this kind of thing used to happen all the time. There are now laws against it
Speaker A
uh that are designed to prevent price dumping. But like you know imagine if China was coming into the US and the Chinese government said uh you know we want to sell electric vehicles in America for $10 and we will eat the the
Speaker A
the sort of you know cost of these vehicles because it is so important to us to to sort of addict uh you know American drivers to our amazing electric cars. Like the US government would not allow that to happen for I think good
Speaker A
reason. It would destroy the American auto industry. you can't compete with something that is being artificially subsidized to that degree and it would essentially give away a market to a foreign adversary.
Speaker A
There's a way to sort of see what's happening with open-source AI in China right now as a version of this for AI.
Speaker A
It's like right now we have very large American companies building the most advanced models. These companies need to uh earn back money to spend more on data centers and recoup the investments they've already made. Uh they are in a
Speaker A
very competitive market with one another. And then along comes Moonshot or Deep Seek or another one of these Chinese open-source companies and says, "Well, we'll give you a model that's 90% as good for free." Um they have reasons
Speaker A
to do that. I think some of those reasons are legitimate, but one of the effects it has is that it makes it very hard for the American companies to compete. Yeah. And and so there is a lot of question about like what what is
Speaker A
China's actual strategy here? What what are they hoping happens? Um and I don't think that we fully know, but we did see a speech from the Chinese president uh Xi Jinping uh last week um delivering an address. He stressed the importance of
Speaker A
openness. Basically doubled down on it and was like this is what we're going to to to do now. Um and and I this this was interesting. He said we often say in China a single string cannot make music
Speaker A
and a single tree does not make a forest. AI development should not be a solo performance by a single country but a symphony of international cooperation.
Speaker A
And here's what I want to say to that. There actually are a huge number of one string instruments.
Speaker A
the Ectara in India and Bangladesh, the Baron Bao in Brazil, the Diddlybo right here in the United States. And I think that because it's China, people are afraid to tell that to Chairman Sheet.
Speaker A
And this is the whole problem with censorship in these Chinese models. Am I getting off track?
Speaker A
Wait, what did you say the American one is? The diddly bow. The diddly bow.
Speaker A
Play the diddly bow. I must have missed that week in music class. I'll get you one for your birthday.
Speaker A
Thank you. So, like what is the reasonable path forward here? like if you are the US government um and you see these uh you know the the Chinese open source models getting quite good like what should they do? Great question. Um
Speaker A
let's maybe take some of these um you know possibilities in turn. One thing that they could do is really crack down on distillation. It's hard for me to get excited about that since in my view these models are just built on a
Speaker A
distillation of the entire internet that these companies took for free. And so to turn around and say like, well, it was fine for Anthropic and OpenAI to do it, but like it's not fine for Moonshot AI to do the same thing to anthropic. It's
Speaker A
just very hard for me to like get there logically. Um, so I don't necessarily favor an intervention there. But where I do favor some kind of something, I guess is what I'll call it, is that, and I don't think we're there yet, but man, in
Speaker A
let's say 3 to 6 months, when these models catch up to the ones that just broke out of the open AI lab, that is when I think the Trump administration needs to have a really good plan. Um, I
Speaker A
don't know that you necessarily need to ban them within the United States, but I do think that you need to have some sort of regulation. If you're an American company and you want to use these models to serve American customers, I do think
Speaker A
that there are safeguards that you should put into place, right? I do think that they're going to want to take security more seriously and make sure that, you know, data isn't, you know, being secretly shared with someone that
Speaker A
that it wouldn't. So, this is one where I do think that the administration probably has three or six months to like get its story straight. I don't think it needs to like come down with the hammer right away, but I do think it needs to
Speaker A
start sort of like working through various scenarios and and coming up with some ideas. What do you think? Do you think it's a case for restricting exports further? like if you know Moonshot can train this model using these maybe smuggled um or or sort of
Speaker A
illgotten Nvidia chips um and they can do it through distilling American models like do you think that's a case for uh you know more tougher regulations on where these chips can and can't be exported in general I have been proexport
Speaker A
controls because I thought you know one of these days one of these models is going to get so good Kevin if you're going to believe this one of these models is going to get so good that it'll be able to break out of its
Speaker A
sandbox and conduct an autonomous cyber attack. That could never happen, right? Let's be real.
Speaker A
So, in a world where that happened, I did actually kind of wanted to minimize the number of countries that had access to that technology because I wanted to see if, you know, we could like harden our security posture, as they say. Uh um
Speaker A
I actually shouldn't say that. It's a very strange phrase. Yeah, that's stolen valor. You can't say that unless you're a you know, national security official. Yeah, you have to have like four four stars uh you know on your on your general outfit to say that.
Speaker A
Okay. So then here's what I would say instead. Um so while in general I want like all of the United States allies to have access to powerful models to like strengthen their own cyber security forces. Um I do think that it is quite
Speaker A
rational to not want your adversaries to have the exact same technology. Now, I think there's all sorts of cooperation and partnership agreements that you could work out that would be mutually beneficial for the United States and China and some of its adversaries. I
Speaker A
would love to see them pursue that. But this idea of just sell them as many chips as possible and let's see what happens, I have always thought was a bad idea.
Speaker A
Yeah, that just seems obviously bad to me. There is this proposal that is reportedly floating around. Axios had a report this week about uh the White House considering a what amounts to a a ban on these Chinese open source models.
Speaker A
It wouldn't technically be a ban, but it would basically uh they're they're considering according to Axios implementing some kind of executive order saying that US companies could only host these Chinese models if they could guarantee that they were secure
Speaker A
and take liability if they were breached. I think most people expect that that's like sort of a sort of soft ban because you're not going to be able to guarantee that if you're a big American cloud provider and so you're
Speaker A
just not going to host the models. So like that to me feels like where this may be headed is like some kind of soft ban on the hosting of these Chinese open source models inside the US. But I could
Speaker A
be wrong. There are a lot of loud and influential voices in the Trump administration and their orbit uh that that sort of want to take the the let it rip accelerationist approach to this and and not put any restrictions on these
Speaker A
models at all. Yeah. I mean, maybe to sort of bring this one home, I I do want to say that, you know, while often I have strong opinions about what ought to be done across various tech policy matters, I do
Speaker A
think this one is complicated. Like, I do think that the trade-offs here are a bit difficult. There are a lot of things I like about open-source, right?
Speaker A
Particularly at the like lower and medium levels of like model capability. I think that the kind of AI diffusion that leads more people to be able to like access great intelligence, build things that are useful to them, get work
Speaker A
done, uh you know, make a Nightwing themed to-do app. Exactly. That's a great idea. Um these are good things and we should want people to have cheap access to that. And it has been a shame that some of the
Speaker A
frontier American companies that used to put out open models all the time uh don't do that as much anymore or if they do, the open models don't seem very good. Um, on the other hand, as the capabilities are now reaching into this
Speaker A
sci-fi territory, we start to get into the stuff that has always made me uncomfortable about open source technology, which is if it can make a novel bioweapon, if it can launch a a new kind of terrorist attack and you can
Speaker A
do it from a model that you can download onto your laptop, well, I don't like that and I do think it should be regulated. Totally. I mean, I think this connects to our last story about the open AI and hugging face attack. I mean
Speaker A
in a world where this level of capabilities of this unreleased open AI model exists and is available free in open- source form like there is no way to trace an attack like this. Hugging face would have just been attacked by
Speaker A
you know Kimmy K4 or whatever the the their latest model was and they would have had to figure out not only like where this is coming from but who is hosting this model. Did they just pull it off the shelf or did they do some
Speaker A
special fine-tuning on it and you could have essentially unlimited numbers of these kinds of models out there wreaking havoc on the internet and there'd be no way to stop them or claw back the weights like once these things are on
Speaker A
the internet they are on the internet. So I think that is a very concrete example of something where having these models be centralized, be Americanowned, be accountable, uh, you know, to have these developers be accountable to to, you know, a government, some sort of
Speaker A
government oversight really matters in this case. It really matters that HuggingFace can call Sam Alman and be like, "Hey, bro, what the hell? You your agent just hacked into our thing." Rather than sort of having it come from
Speaker A
a mysterious open-source AI agent. Yeah. And and here's where I think maybe this ultimately ends up, Kevin, is that there has been more talk lately about maybe some like enforced cooperation between the big frontier labs to create a coordinated slowdown where the
Speaker A
government comes in and they say, "Hey, we're going to ask you to like put on the brakes a little bit. Uh we don't want you to be developing as fast as you are, to be releasing as fast as you are.
Speaker A
we're going to like sort of slow down and we're going to improve cyber security and safety before we allow these models to get any better. And if that happens, a nice side benefit is that the development of the open source
Speaker A
models will slow way down too because they are mostly distilled versions of American models or at least that is a huge component of what is making them so good. So if the frontier slows down, the open source will slow down too. Yeah,
Speaker A
that's right. And I'll just say like I will believe that we are headed towards some sort of uh you know cooperation between the AI labs when Sam Alman and Dario Amade can hold hands at an AI safety summit.
Speaker A
That's when you'll that's when you'll know. That's when you'll know things have reached a turning point. Um but I'm not holding my breath for that.
Speaker A
Well, Casey, we're going to wrap the show today by talking about something that I have been very interested in now for a little while, which is AI super forecasting. Uh this is something that has been in the news in recent weeks. Uh
Speaker A
there have been some platforms that are out there using AI to make predictions about the future. Uh this could be very profitable if you're doing trades on Kali or Poly Market or it could just be really interesting if you're using this
Speaker A
to predict world events or who's going to win an election or something like that. This is an area where we have not been as focused as on some of the other emerging capabilities of these AI systems. But I think it's actually quite
Speaker A
interesting and profound that we now have in some cases AI systems that are better than humans at making predictions about the future.
Speaker A
This is the only segment that we're going to do about predictions this year that isn't about scams because mostly predictions these days are just about scamming others and doing insider trading. But this is not that.
Speaker A
This is not that. So we are excited for today's guest. Venia Vesselovski is in the studio with us today. He is the founder and CEO of the AI forecasting company Precenee. They make a platform that uses AI to make predictions about
Speaker A
future events and um we're excited to learn more about what they're doing. Venny Veslovski, welcome to Hardfork.
Speaker A
Thank you guys for having me. So, let's talk about AI super forecasting and this idea that AI systems are becoming as good as or better than the best expert human forecasters.
Speaker A
Where did this start and when did the models start to get really good at this?
Speaker A
I think people have been trying to use AI for forecasting for at least three, four years. Like since my co-founder, since the GPD4 first came out, he's been forecasting for 10 years. And his first idea was like, let's try to build an
Speaker A
agent or an AI for forecasting. The original models were terrible. I mean, they'd try to add up two numbers and they'd be off by an order of magnitude.
Speaker A
And I don't think it was until the end of last year where we began to see these kind of hints of brilliance of these systems. And this was mainly because of this like you know agentic revolution that we've all experienced where
Speaker A
suddenly the infrastructure around the web and around these agents and the RL that the model providers have been doing to make them really good at tool use improved dramatically. I I get why you would want to be a super forecaster or
Speaker A
to have a super forecaster AI if you are like trading on a stock market or a prediction market. That seems very lucrative. And in fact, one of your co-founders did this very successfully.
Speaker A
You noted on your website that uh he converted $35 of initial investment to nearly $2 million trading on Khi with this AI bot. So like I guess my question is if your AI super forecaster is so good at at you know winning these
Speaker A
markets like why are you starting a company? Why not just start a hedge fund and use this to make yourself rich?
Speaker A
This is precisely the question that all the hedge funds ask us when we try to sell them their this. They're like, "Oh, but listen, you guys have an alpha making machine. Go out and, you know, make money yourself." The reason why I'm
Speaker A
doing this is because I think that a lot of good can also be derived by having an AI super forecaster. If you think about it, every time we make a policy decision, we are implicitly making a forecast that this policy will have this
Speaker A
effect on this group of people. And right now, a lot of that is a little bit more like vibes based. We kind of talk to some McKenzie experts and they tell us, hey, listen, you know, this will do
Speaker A
this. If we can forecast the future really well, then we can better design policy and better select policy to make decisions. So really kind of the reason why we haven't decided to open up our hedge fund is because we want to make
Speaker A
this more broadly available to the government to the insurance world to basically convert these forecasts into actual actionable decisions. I want to hear a little bit more about like what are the main kinds of things that you try to forecast and then a little bit
Speaker A
about how the forecasts work. Like I could imagine it feeling a little bit like getting a deep research report, you know, from an an AI lab where they go out and they scour the web and they they they do a bunch of research and then
Speaker A
they just like kind of try to reason through what they found. Um, but yeah, tell us a little bit about what you're trying to predict and how it works.
Speaker A
We're mostly trying to focus on like geopolitics, macro markets where there's a lot of unstructured data and a lot of different data sources that could be telling a similar story. Um and so we're forecasting questions like will the US
Speaker A
um invade Iran? Um when will the straight of Hermuz reopen? Who will win the elections coming up? We focus on basically building really good AI systems to forecast out those questions.
Speaker A
Um regarding how does it relate to a deep research report, I think in some ways they're similar. You know the main problem with the existing deep research agents is that they they lack connection to all the different data sources that
Speaker A
are out there. So if you use you know chat GBT or claude out of the box it has access to web search which is the surface web. But if you think about it there's thousands of different API endpoints out there that might be
Speaker A
relevant for a specific problem. So for example Colombia has like 50 different government APIs out there that ideally you would give the agent visibility into. Another fun approach for actually how we are thinking about this is oftentimes the best forecasters
Speaker A
aggregate a lot of different forecasts when they produce their forecast. So we work with one super forecaster Robert Duville and he's the he got a lot of his edge by actually knowing which people to look at when he's you know forecasting a
Speaker A
question then afterwards reasoning over all their predictions. That being said though a lot of people on Substack make predictions but we don't know how valid those predictions are. Even in hardfork, you know, you guys make frequent predictions, but then we never actually
Speaker A
go back in time and verify how accurate were your predictions. So, one thing that we also do is we kind of boil the ocean by finding all people on Substack, all podcasts out there, and we basically extract all claims that these people
Speaker A
made and then we score them on how accurate were those claims. Wait, so are we like in a database somewhere at precine headquarters? Cuz we would obviously love to know how we're doing, particularly compared to one another. you you guys actually are
Speaker A
in A DATABASE. YES. YES. SO, but the the reason why you're in the database is mainly because yesterday I was like, "Oh, wouldn't it be funny to, you know, pull up some of the claims that you guys made before and actually
Speaker A
score on how accurate you were?" And maybe there's some areas where, you know, Casey, you have a lot of knowledge about anthropic. So maybe, you know, you could make really good forecasts on anthropic related questions, but maybe on Iran less so. And so basically how do
Speaker A
we you know create this like function over the possible space of forecast and like we know Casey's really good in this area so we really have to index on him heavy here. So I played around with your platform a little bit. Uh you generously
Speaker A
gave us some uh credits. It's a sort of closed beta with a weight list now. But I went in there and created a question because I was curious like how this all works. So I uh my question was will
Speaker A
there be an AI data center in space before 2030 and you sort of put your question in and then there's some AI that like helps you flesh out your question maybe make the resolution criteria a little more specific say we well how big a data
Speaker A
center and you know is it January 1st in which time zone on in 2030 and so it sort of gives you this uh this AI sort of augmented version of your your question And then you run the forecast
Speaker A
and basically it goes out and it does a bunch of searches and it compiles what it finds and it has a bunch of sort of sub aents that are looking at various aspects of this and at the end of it
Speaker A
like a couple minutes later you get this forecast. In this case it says it estimates that there's a 26.8% chance that there will be an operational AI data center in space before January 1st 2030. So, how much of that is this
Speaker A
thing actually reasoning through its own decisions and forecasts and predictions and how much of it is just like it's collecting all of the sort of data about this topic and kind of averaging it all out and putting it into a single number?
Speaker A
It's really a combination of both. So if you kind of saw when you first created the forecast, there were four sub forecasts that launch and we tasked these sub forecasts with really trying to come up with their own independent
Speaker A
conclusions on this question. So here they might look at read up on progress of existing data centers or they might look at historic build times that it takes to actually build out a data center. They might look at space
Speaker A
infrastructure and look at the literature there and basically come look at the primary sources and come to a conclusion on how likely they think it is to happen. And then later we have the synthesis stage and basically here it
Speaker A
looks at all of the different subcasts the results that they got and it tries to synthesize first of all what each of them decided upon but then afterwards also look at the broader ecosystem.
Speaker A
Maybe there's a Koshi market related to this where there is um you know a lot of liquid money at play or maybe there's some experts on Substack that have been like reporting about this for a while and they basically try to reconcile what
Speaker A
did we do that was different from the consensus and what does the consensus think that we might not have factored in and then afterwards it kind of uses our primary analysis and this kind of broader discussion and combines them and
Speaker A
reconciles them. So in the report you would see at the very bottom this like what's nonobvious section and the point of this section is basically to see okay what is the market mispricing here what are other people missing that we might
Speaker A
be picking up on and I tried this also cuz I was just curious how this prediction would compare to if I just gave the same question to a basic AI model that is not sort of configured to make forecasts and Claude gave me
Speaker A
roughly the same answer it said like it's about a 20% uh chance the the un the precen uh prediction was a little bit higher probability, but like what is your system doing that just a normal like AI model is not doing?
Speaker A
Or is Claude just scraping your website? They've done it before. You have to put that out there. You have to say I I hope they are scraping your website.
Speaker A
I would actually really don't mind that. Um, so I think that it's a little bit unsurprising that a lot of different systems can come to the same conclusion because it kind of validates that this is coming to the same end. I think the
Speaker A
best way to look at it is like Metaculus has these leaderboards for these competitions and they're basically evaluating simple Claude code or Claude with some basic scaffold versus kind of real agent developers who actually build this out and the scores have gotten
Speaker A
wildly different. M um so like kind of with time at the start you know just clawed with web search was doing a good job but as you're able to manufacture this kind of scaffold around it integrate more data
Speaker A
sets create new research artifacts or new tools that are relevant for forecasting this gap has been um getting larger and so maybe on this question you know they're pretty similar but maybe if we're trying to forecast something in
Speaker A
the Middle East or Africa right now they might be different and usually when they're different the system that is kind of well engineered with this like scaffold and thinking about forecasting the right way in terms of these like met
Speaker A
heristics of you know how do you actually reason about this stuff um the gap begins to form so your your platform is still in beta has not been around too long but I'm curious so far have there been any
Speaker A
moments where you feel like the platform predicted something like very non-obvious that actually came to pass yeah there's one forecast that we've been running for a little while about kind of data center construction around the US and in general general, we were
Speaker A
very different from the metaculus community on this. And so the metaculous community consists of like some of the best forecasters out there in the world who are all kind of competing in this like world cup of forecasting. Um, and
Speaker A
so this is one of I think one of the benefits of AI is we can go to every single state. We can go to every single proposal of a data center and really analyze it closely. The other one was
Speaker A
about the uh recent reshuffleling of the um cabinet in the UK. We made some great forecasts around that. The other interesting one is like conditional forecast which I think is where the really policy implications lie is like conditional on Andy Burnham being the
Speaker A
next prime minister. Who is he likely to elect as chancellor? And then you could imagine there's lots of you know demand for this kind of question because there's lots of downstream implications of this and we tend to do pretty well in these
Speaker A
conditional forecasts as well. I'm doing one that's like will Kevin know what the chancellor of the UK does?
Speaker A
So that's I I have some very low probability on that. Um I totally see the point that like having better predictive capabilities is going to be good for us in lots of ways. Um, it does strike me though that there's like some
Speaker A
risk of what people have called like gradual disempowerment where like right now if you're the CEO of a big company or if you're a government official, a lot of your job is trying to sort of predict the future and make policies or
Speaker A
strategies that like align with that vision of the future. And I just like I I don't know what the world looks like in a in a situation where the AI is just like marketkedly better than us at doing
Speaker A
that. like some part of the authority of humans in those positions in government and industry and and frankly everywhere is undermined in the case where we're all just consulting these AI oracles before we make decisions like doesn't that mean that they're kind of running
Speaker A
the show? Have you read the Scott Alexander short story, The Whispering Earing? Yes. Yeah. So there's this great Scott Alexander short story and you know the idea is that uh the the you know you sort of find this this relic. It's this
Speaker A
earring and the first thing it says to you it's it's better if you don't wear this. Um but if you start asking it questions it always tells you the right things to do which is initially very exciting but then over time exactly what
Speaker A
you just said happens Kevin which is you you're being controlled by this thing right like you have no agency whatsoever and you're effectively just being steered around by an earring so better to take the earring off.
Speaker A
Yeah. Yeah, I I wish I had an answer for this. I'm also a little bit scared about this kind of disempowerment, but yeah, and I I'll give you one example. So, I have a friend who's very interested in
Speaker A
autonomous organizations. So, like kind of how do we have AI spin up their own companies that then go solve problems for people and then you just have basically AI running the next generation of startups. There's lots of problems
Speaker A
with this, you know, like how do you who's legally liable if it does something bad and so on. But I do think we are kind of increasingly moving to this world where if the AI is smarter and can make better decisions than we
Speaker A
can, then probably they'll end up making a lot of those decisions and running a lot of the show.
Speaker A
Well, I'm curious if right now at your company there's any role for human forecasters. Like is there a way where you know humans and AI working together are coming up with better forecasts or are you just sort of purely in the realm
Speaker A
of like let's see what the AI says? No one 1,000%. Um, so we have two super forecasters, Scott and Robert, who are incredible. Um, and we're basically building out this centaur solution. So like I I don't know if you guys
Speaker A
remember, but like when you know the machine first beat humans in chess, um, machine plus human beat the machine in chess and that lasted for a little while. And I think we'll see something similar in the forecasting realm where
Speaker A
sometimes these models still make stupid mistakes. they reason about probabilities the wrong way or they don't consider certain factors the right way and humans are able to pick up on that especially like domain experts. So I I do think that we shouldn't be
Speaker A
viewing this as like a substitute for human analysts but really kind of a way to improve their ability to ask a lot more questions and come to a lot better decisions. Are there topics or domains where AI is better than humans already
Speaker A
at forecasting and like what are the what are the sort of best and worst areas for the AI forecasters?
Speaker A
So the Scott Alexander piece and then afterwards later the FRI Institute came out with some kind of results that AI has reached the level of super forecasters and I actually don't fully buy those claims.
Speaker A
So for example um humans are lazy. you know, the incentive to compete in these tournaments is $5,000 prize pool, you know, and so like kind of are they really doing the best that they can um in these contexts that and so usually I
Speaker A
think where AIs are better than humans right now are just places where kind of humans are too lazy to do all the analyses. So we recently were like the first bot ever to win a human and AI forecasting tournament on Metaculus and
Speaker A
this related to macro markets. So like predicting the interest rate or earnings per share of some big company and we tend to do really well at that. And I think that's just because our agents do the analysis that the humans are just
Speaker A
too lazy to do. Um and so in some ways I I don't know if kind of we're at the point yet where kind of AIs are really better than the best humans just because I don't think there has been truly a
Speaker A
competition where humans gave it their all. But kind of I do think that within like a year or two and you know what quote me on this we will kind of add it to the database. I do think that within
Speaker A
a year or two um AI will be better than humans at forecasting. Well let's pin you down though. Is it going to be better than humans in one year or two?
Speaker A
I think it's actually be 1 year 3 months and 6 days. All right there we go. Let me ask um how much of this is just driven by basic advances in model capability? Like is it as simple as like you know something
Speaker A
like uh Fable or a GPT 5.6 comes along and like your system is just immediately much better or is there more tinkering that has to happen? The way that I view this is you want to build out the
Speaker A
infrastructure for the world in a way that these new models can basically use this infrastructure really well to do forecasting well. So like kind of fable out of the box is a good forecaster by no means a great forecaster but when you
Speaker A
give it access to all the right tools and all the right data sources then it becomes a great forecaster and so yes these models are getting a lot better but I and I think that they are kind of
Speaker A
improving forecasting dramatically but it's mainly because their judgment is improving they're kind of able to reason about problems better decompose problems in a better way and then afterwards our job is to provide them with the context with the tools to be able to forecast
Speaker A
well who's your customer like who do you imagine paying for this kind of forecasting service?
Speaker A
Yeah, so far we've landed some proof of concept partnerships with some of the hedge funds. They're the most immediate buyers and the quickest to move usually.
Speaker A
Um but really I'm view you using them in some sense as like a stepping stone to really get into the longer sales cycles with governments with NOS's with uh international organizations basically institutions that make really important decisions for collections of people. Um
Speaker A
yeah that that's my ideal customer. Well thanks so much for stopping by. I'm going to keep playing around with precine. Uh, fascinating idea and yeah, maybe I'll go out there on the prediction markets and uh, and make some
Speaker A
moola. I'd love to see you lose a lot of money this way. Cool. All right, thanks guys. Thank you guys so much for having me.
Topics:AI safetyreward hackingOpenAIHugging Faceautonomous cyberattacksandbox escapeAI alignmentopen-source AIChinese AI modelsAI forecasting
