# Why Frontier AI Labs Could Go Bankrupt | Ege Erdil

Markdown copy of https://realalpha.app/v/00iQmHkfYx8/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

Ege Erdil argues that frontier AI labs face risky valuations because heavy compute spending may not translate into durable margins.

- Show: MTS
- Hosts: [Theo Jaffee](https://realalpha.app/people/theo-jaffee/) & [Sophia Dew](https://realalpha.app/people/sophia-dew/)
- Guest: [Ege Erdil](https://realalpha.app/builders/ege-erdil/) is a research engineer at Google DeepMind; former Mechanize cofounder and Epoch researcher.
- Published: Nov 12, 2026
- Length: 31 min

## The best minutes

- [0:59](https://realalpha.app/v/00iQmHkfYx8/?t=59) [Frontier labs may struggle to sustain high margins](https://realalpha.app/s/frontier-labs-may-struggle-to-sustain-high-margins/) (5 min)
- [9:36](https://realalpha.app/v/00iQmHkfYx8/?t=576) [Long-term compute commitments could put lab equity at risk](https://realalpha.app/s/long-term-compute-commitments-could-put-lab-equity-at-risk/) (3 min)
- [16:20](https://realalpha.app/v/00iQmHkfYx8/?t=980) [Compute and data constrain how fast AI research can accelerate](https://realalpha.app/s/compute-and-data-constrain-how-fast-ai-research-can/) (3 min)

## Also worth watching

- [21:05](https://realalpha.app/v/00iQmHkfYx8/?t=1265) [AI timeline forecasts need measurable real-world outcomes](https://realalpha.app/s/ai-timeline-forecasts-need-measurable-real-world-outcomes/) (5 min)
- [12:43](https://realalpha.app/v/00iQmHkfYx8/?t=763) [Profitable memory stocks look safer than frontier labs to Erdil](https://realalpha.app/s/profitable-memory-stocks-look-safer-than-frontier-labs-to/) (3 min)
- [6:09](https://realalpha.app/v/00iQmHkfYx8/?t=369) [Low switching costs weaken AI labs’ pricing power](https://realalpha.app/s/low-switching-costs-weaken-ai-labs-pricing-power/) (3 min)
- [25:34](https://realalpha.app/v/00iQmHkfYx8/?t=1534) [Erdil sees the METR report as weak evidence of AI acceleration](https://realalpha.app/s/erdil-sees-the-metr-report-as-weak-evidence-of-ai/) (2 min)

Skip the sponsor reads, especially the mid-episode break and post-interview VCX ad.

## Full analysis: Ege Erdil puts lab bankruptcy risk above 10% as an example

- Erdil called lab bankruptcy a decent chance, giving more than 10% as an example over three years.
- His concern is long-term compute commitments meeting uncertain demand, alongside weak customer switching costs.
- Operators should map contract exposure and test whether model portability limits pricing power.

Its 6 sections, what it means for you and what to watch are on [the page](https://realalpha.app/v/00iQmHkfYx8/#analysis), free with an account:

- [1:27](https://realalpha.app/v/00iQmHkfYx8/?t=87) Technology importance does not guarantee durable margins
- [5:02](https://realalpha.app/v/00iQmHkfYx8/?t=302) Labs must keep spending to defend model leadership
- [8:40](https://realalpha.app/v/00iQmHkfYx8/?t=520) Easy switching limits model providers’ pricing power
- [10:17](https://realalpha.app/v/00iQmHkfYx8/?t=617) Compute commitments can turn weak demand into equity risk
- [13:35](https://realalpha.app/v/00iQmHkfYx8/?t=815) Current profits offer a firmer starting point
- [21:30](https://realalpha.app/v/00iQmHkfYx8/?t=1290) Forecasts need observable outcomes and disclosed evidence

## Summary

Ege Erdil is a research engineer at Google DeepMind. He was previously a cofounder of Mechanize and a researcher at Epoch.

Erdil argues that trillion-dollar valuations for frontier labs assume they can turn important technology into durable high-margin businesses, but their compute spending and weak switching costs make that uncertain. He says long-term compute contracts could put lab equity holders at risk if demand or compute prices fall, while preferring memory stocks and remaining bullish on Nvidia. He rejects claims of a near-term intelligence explosion, arguing that progress remains constrained by compute and data and that forecasts should specify observable outcomes. He estimates full automation of white-collar work is roughly 20 years away and calls recent evidence of AI R&D acceleration weak.

### Minute by minute

- [0:59](https://realalpha.app/v/00iQmHkfYx8/?t=59) **Frontier labs may lack durable margins.** Erdil says the labs’ valuations assume they can turn important technology into durable high-margin businesses. He compares them to airlines: the technology can matter enormously while the businesses still require high capital spending and produce low margins. He says durable high-margin businesses usually have a moat and low ongoing costs.
- [6:09](https://realalpha.app/v/00iQmHkfYx8/?t=369) **AI may not be winner-takes-all.** Erdil says labs must keep investing to maintain the lead, and he questions whether scale gives them a decisive advantage. He notes that Anthropic has not pulled ahead as much as its resource gap over Chinese open-source labs—perhaps 30 times more compute—would suggest. He also says customers can switch AI models more easily than cloud providers.
- [9:36](https://realalpha.app/v/00iQmHkfYx8/?t=576) **Compute contracts create failure risk.** Erdil says labs could fail or restructure while their compute assets remain, wiping out equity holders. He estimates Anthropic’s commitments could approach $1 trillion, but cautions that he is unsure of the exact number and timing. He puts the chance of lab bankruptcies over the next three years above 10% but below 50%, citing the risk that falling compute prices leave long-term contracts underwater.
- [12:43](https://realalpha.app/v/00iQmHkfYx8/?t=763) **Erdil favors memory stocks and Nvidia.** Erdil says memory stocks look safer to him because they are already high-margin businesses, provided demand stays strong. He calls OpenAI riskier than Anthropic. He also agrees with a $15 trillion fair-value estimate for Nvidia, arguing that its revenue can keep growing and that recent market sell-offs may overreact.
- [15:53](https://realalpha.app/v/00iQmHkfYx8/?t=953) **Compute and data constrain AI timelines.** Erdil rejects the idea of a near-term intelligence explosion, arguing that automating AI development still depends on data and compute. He says cognitive labor has diminishing returns once those inputs become bottlenecks. He points to correlations between compute scaling and software progress across fields including chess, SAT solvers, natural language processing, and computer vision.
- [21:05](https://realalpha.app/v/00iQmHkfYx8/?t=1265) **AI forecasts need testable outcomes.** Erdil says people predicting an intelligence explosion should specify observable events that could prove them wrong, rather than relying on abstract metrics. He cites a prior forecast that AI would execute machine-learning experiments from high-level instructions within about three years as the kind of testable claim he values. He says predictions about robots also need clear definitions.
- [25:34](https://realalpha.app/v/00iQmHkfYx8/?t=1534) **Erdil calls METR evidence weak.** Erdil says the METR report offers weak evidence for accelerating AI research: he questions what its progress metric measures and notes that some supporting evidence is redacted. He also argues that social pressure can shape how people update their views, and says he has not seen decisive evidence in either direction.
- [27:51](https://realalpha.app/v/00iQmHkfYx8/?t=1671) **White-collar automation may take 20 years.** Erdil revises his earlier estimate of 30 years to roughly 20 years for full automation of white-collar work. Asked what to put to Buck, he suggests asking for a concrete forecast; he says a tenfold rise in global primary consumption over the next three years would make him admit he was wrong.

## Transcript

MTS, hosted by Theo Jaffee and Sophia Dew, with Ege Erdil.

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

[0:00](https://realalpha.app/v/00iQmHkfYx8/?t=0) For Anthropica, I think it's close to a trillion dollars that they have committed over the next three years. And that's a very large amount. Like Google's annual revenue is only in order of like $500 billion. That's revenue, that's not even like profit. So if you are making a spending comment of like $800 billion over 10 years, like that's basically like you're saying I'm going to spend Google's entire income for 10 years. And I think that that is a real danger because labs have to enter into these long-term, multi-year contracts for compute. But it means the equity holders could get wiped out.

[0:44](https://realalpha.app/v/00iQmHkfYx8/?t=44) All right, we are back. We are live with Ege Erdil, who's a research engineer at Google D-Mine, previously co-founder of Mechanize and a researcher at Epoch. We love Epoch. Ege, welcome to MTS. Yeah, thanks for having me. so you have a lot a lot of takes a lot of things to talk about um one thing i'm definitely interested in is lab valuations so you seem to be more bearish than most on lab valuations you know opening eyes just raised at 1.25 trillion um anthropic is supposed to ipo at something between roughly 1.3 and 2 trillion so do these lab valuations make sense at all i think they're too high uh like i personally wouldn't invest in these companies but i think the reason i would say that is because i just don't see the the i have never heard a reasonable story for i've they become like sustainable high margin businesses i think there's a thing going on in the markets where people think the technology is going to be a very big deal which i think is true and they're extrapolating from that saying and therefore these companies that are building this technology are going to be very valuable companies but uh like a for example air like airplanes were a very big technology and uh you know all around the world people spend a lot of

[2:05](https://realalpha.app/v/00iQmHkfYx8/?t=125) money on flights and so on and they become a very big deal but airlines are very bad businesses and airplane manufacturing is also very bad business like it's like a high capex business where your margins are very low and uh if you just look at the valuation of these companies they're very low for that reason so i think like that's that's my main worry about these companies i don't know like airbus stock has done quite well boeing stock had done quite well until the 787 max disaster uh and covid but like it seems like well it's different what is the valuation ah valuation yeah um let's find out airbus is 166 billion uh euros which is 184 billion usd And then Boeing is, what is Boeing?

[2:57](https://realalpha.app/v/00iQmHkfYx8/?t=177) Probably something similar. Boeing market cap, $138 billion. Yeah. Do you think, like, why do you think lab valuations, like, why do you think labs will not be able to be sustainable high margin businesses? is the idea that like what is the argument like the default argument should be that no business can be a sustainable high margin business sustainable high margin business are very rare so the default state of any business is to end up like like if you look at the average profit margin over all business in the economy or something like that that's like quite low so what are the most high margin businesses so yeah so like most high margin businesses are ones that can have like low cost, low ongoing cost, they have some kind of advantage or moat or economy of scale or whatever, that makes it so that they can maintain their position without having to spend a lot of money.

[3:59](https://realalpha.app/v/00iQmHkfYx8/?t=239) try to compete with, like, new entrants coming into the market. Like, I think a good example of this for actually a very long time was Google. Because they had this search system, which was, because of network effects, it's very difficult for someone else to come in and compete with a new search engine. And then that can be a low-capx business that's sustainable over long term. Like, some of the hits Google's valuation has taken recently is because people have become worried about that. and they think, for example, something like the cloud revenue is less high quality because you're going to need a lot of CapEx to sustain it. A lot of the big tech companies are good examples of businesses that were, in this sense, good businesses because they could have low CapEx, they could throw off tons of free cash flow and pay to do stock buybacks and return a lot of money to shareholders because of that. And they didn't have to have this ongoing incredible amount of capex to maintain their position but for the ai labs it seems like they do have to have a lot of capex to maintain their position like if they stopped spending yeah like the only reason yes like the anthropic can charge very high gross margin on inference they can charge like 85 90

[5:07](https://realalpha.app/v/00iQmHkfYx8/?t=307) gross marginal inference but the reason they're able to do that is because they have the best model and if you if they wait for one or two years then they won't have the best model anymore so they won't actually end up making that much money.

[5:19](https://realalpha.app/v/00iQmHkfYx8/?t=319) **Ege Erdil:** Their position is entirely dependent on them staying in the lead, which is a very expensive thing to do.

[5:26](https://realalpha.app/v/00iQmHkfYx8/?t=326) And so far, you could have told a story about how as the costs get so high to be at the frontier, then there will be economies of scale, which would mean the person who was ahead is going to have a lot of computer, going to have a lot of capital, and they will be able to outcompete other people just by virtue of that advantage. But if that story was right, then you should expect these labs to be pulling way further ahead of labs from China with way, way less resources than they have. And that has not been happening. I don't think they've been pulling ahead. So I think that's also concerning. There's evidence that their business model is less good than you might have thought.

[6:08](https://realalpha.app/v/00iQmHkfYx8/?t=368) There's definitely evidence against it because if you were seeing the opposite, it would have been evidence for it. So conservation of expected evidence. you know we were like talking earlier about how this post that was making the claim that ai was like a winner takes all market uh supposedly pulled from an anthropic investor meeting yeah so i'm guessing you probably find that a bit silly but i want you to hear your your thoughts i mean i don't see the argument i think the argument is something like anthropic continues to grow revenue more quickly than it grows compute cost um which allows it to grow its margins and that AI will allow it to accelerate its own research more quickly than other companies and produce better models more quickly.

[6:53](https://realalpha.app/v/00iQmHkfYx8/?t=413) That's not happening right now. So they are not pulling ahead of open source labs right now. So it could have way less resources. Open source labs in China have maybe 30 times less compute than Anthropoc does or something. It's like a huge difference. So you should be seeing that make a bigger difference if that was the case i don't think it's making us a big enough difference well there's also the argument that um more intelligent models like the most intelligent models are worth um more than linearly more than less intelligent models because uh new types of applications get unlocked that weren't possible before and people are willing to pay more for this yes i agree people are willing to pay more for the most intelligent model i agree with that but like that that i don't think that means you can have like a like a very high margin business on top of that right yeah well like i guess to ground it and like just like software companies like have they been able to remain high margin relatively sustainably or like how has that looked through time? I think for the software companies, the ones that can remain high margin sustainably are the ones that have some kind of lock-in or network effect or high switching costs from that.

[8:10](https://realalpha.app/v/00iQmHkfYx8/?t=490) So, for example, I think the reason cloud has been able to remain relatively high margin, like typical margin for cloud providers used to be a 30% or something like that. And I think the reason that was sustainable was because cloud switching costs are really high. So if you've built your business around like a specific cloud like AWS, migrating to GCP is extremely annoying and difficult and you don't want to deal with that. So that gives AWS a bunch of pricing power to charge high prices. And that's what I think has allowed cloud to be a longer

[8:40](https://realalpha.app/v/00iQmHkfYx8/?t=520) **Ege Erdil:** term sort of high margin business. But switching costs for AI are really low. So you can very easily just switch to a different model. It's completely trivial. And as a result of that,

[8:50](https://realalpha.app/v/00iQmHkfYx8/?t=530) I think the ability of people to charge high margin is much weaker. At least that argument doesn't work. So you have to say I'm talking a different argument. And I just don't see the argument. Yeah. 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.

[9:29](https://realalpha.app/v/00iQmHkfYx8/?t=569) 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. Lovable.dev. Now, onto the episode. So when you say there's a decent chance of a reckoning in the next two to three years with the labs, what does that mean concretely? Will they go bankrupt? I think it could like go bankrupt yeah like cease to exist yeah or just have to restructure or yeah i mean i think that the assets would not go away right i mean they have a bunch of compute uh the date that they have or they are entitled to as a result of relationships with other people it's not like going bankrupt means that goes away but it means the equity holders could get wiped up and you know and i think that that is real danger because labs have to enter into these long-term, multi-year contracts for compute. For Anthropica, I think it's close to a trillion dollars that they have committed over the next three years. And that's like a very large amount. Like, for reference, I think like Google... Isn't it like the next 15 years, though?

[10:35](https://realalpha.app/v/00iQmHkfYx8/?t=635) I'm not exactly sure if it's the next 15 years. I think there will be very large commitments even over the next three years. I don't know the exact number. And with OpenAI, I don't even know because I don't think it's very transparent. um like i don't know what they have committed to but if you uh if you uh like i think this is because this is the only way they can actually get the other people who are doing this capex directly to have the reassurance of wanting to go through with it so they have to say look we will like buy like whatever you can produce so you don't have to worry about that but that puts them on the line for like a very large amount of spending, I think like Google's annual revenue is only in order of like $500 billion. That's revenue. That's not even like profit. So if you are making a spending comment of like $800 billion over 10 years, like that's basically like you're saying I'm going to spend Google's entire income for 10 years. Yeah. Like that's like a huge commitment. And it's also a bet on compute prices. Because if things go south, let's say, and then compute price almost certainly going to fall, especially for like older GPUs, which the demand has been holding them up. And they've been going up over the past, I don't know,

[11:53](https://realalpha.app/v/00iQmHkfYx8/?t=713) **Ege Erdil:** year or so. But if for whatever reason, the demand for AI just doesn't materialize as much as we think, then compute prices fall dramatically, then all of these commitments are underwater. And it's essentially like you've committed to buy like an NH100 hour in two

[12:11](https://realalpha.app/v/00iQmHkfYx8/?t=731) years for $3, but it's falling to like $1. And then your margins are going to be terrible because everyone else will be just buying the compute for way cheaper and like serving things for way cheaper than you are. So it's a very dangerous situation. Like it's like a very levered bet that this will go well. And they could go well. I'm not saying that it's inevitable that they will go bankrupt or something. I think them going bankrupt still remains below 50% chance over the next three years. But I think there's a decent chance of it. I think more than 10% chance, for example. And so it's very risky.

[12:51](https://realalpha.app/v/00iQmHkfYx8/?t=771) And I think they're aware that it's very risky. Right. Right. So what kinds of trades would you make on AI right now? Financial advice. Okay. Well, I mean, I can interpret that question in different ways. Because personally, I wouldn't really do very much. But that's due to my personal circumstances, which I don't necessarily want to go too much into. Basically, I have large positions that are correlated with AI that I can't necessarily get out of immediately. So for me, I will think about it differently. But if I had just some amount of money and I had to allocate it, then I think the companies I would invest in are companies that are making a lot of money right now but that are not valued highly.

[13:35](https://realalpha.app/v/00iQmHkfYx8/?t=815) **Ege Erdil:** So I think, like, if you want to make it better in my eye, like, memory stocks look way better to me than the labs.

[13:42](https://realalpha.app/v/00iQmHkfYx8/?t=822) Like, I think memory stocks don't have to tell this weird galaxy brain story about how eventually, you know, they're going to become high margin. Like, they're already high margin businesses right now. Like, they don't need to tell any story. Like, the entire hesitance is, like, whether they can sustain it. But if demand continues being very high, then they will sustain it. And then they're way safer, I think, as bets than the labs. Labs could way more easily just completely blow up. OpenAI more so than Anthropic, I think. OpenAI more risky than Anthropic. But even Anthropic is very risky. Well, what about NVIDIA?

[14:13](https://realalpha.app/v/00iQmHkfYx8/?t=853) You've tweeted about being an NVIDIA shareholder. You said the fair value of NVIDIA is $15 trillion, which would imply that it's substantially undervalued still. I think that's right. Why do you think that? I mean, I think just the revenue, I think, is just going to continue to grow. And I think the arguments, like, for example, the recent sell-off in NVIDIA, right? I just don't see, I think that's just an overreaction. I think, like, often the market just overreacts to this kind of news, which don't really tell you very much about the long-term viable in the business. The same thing happened with, like, Google's recent early call, where I think there was a 10% drop in Google's share price or something.

[14:51](https://realalpha.app/v/00iQmHkfYx8/?t=891) based on factors that I personally don't think have much relevance at all on the long-term profitability of business. Like it's just like often these swings happen and like then my recommendation would be that if you think it's like unjustified, then you should just buy more and then in the long term you will do better. But like my argument is just like very simple. Like my argument is that if you can have a high margin business right now, like you already have a high margin business, then it is very strong evidence that it's possible for you to have a high margin business in the future as well.

[15:22](https://realalpha.app/v/00iQmHkfYx8/?t=922) You can tell some argument about, oh, how margins are going to go down in the future. They probably will if they're currently historically high. But, you know, like, you're not in the position of someone who has never been profitable, who is trying to make an argument for how in the long run they're going to have a high margin business. It's a much harder sell. So I would just anchor way more on that and sort of this objective evidence and less more about stories about, like, how the labs are going to be profitable in the future, which to me still seem quite speculative and quite vague. Right.

[15:53](https://realalpha.app/v/00iQmHkfYx8/?t=953) So one other thing that I want to talk about is timelines, right? You had this famous debate with... Daniel Cucatazzo and Ajay Khotra back in 2023, where you had much longer timelines than either of them. Now, Ajay has updated substantially towards very short timelines. Same as Daniel, they put out this blog post yesterday saying it looks like the intelligence explosion will happen in 2027. You disagree with this. So a couple of questions are, why do you disagree? Where are the cruxes? And also, when do you think the intelligence explosion will happen, if at all? I don't think it will happen is the answer to the second question.

[16:33](https://realalpha.app/v/00iQmHkfYx8/?t=993) I think it's like not a good concept to be talking about personally. But I think for the first question of like, why do I disagree with them? I mean, I think that I think with Daniel, it's just very clear that he has he just thinks we'll have a software on the singularity. Like, like, that's that's the key. I don't think that's going to happen. He thinks it's going to happen. And then you can try to get into why he thinks that. And I think if you look at the models, then I just think there's way more that you have to do to automate, to completely automate the process of making AI better.

[17:07](https://realalpha.app/v/00iQmHkfYx8/?t=1027) **Ege Erdil:** And he doesn't, oh, sure, for example, I think a lot of the progress is just driven by data and compute.

[17:12](https://realalpha.app/v/00iQmHkfYx8/?t=1032) And then he agrees with that, but then he doesn't think it's going to be a bottleneck. So the difference is that he thinks you can compensate for this by just having a lot of cognitive labor going into the process of making AI better. And I just don't think that. I think the returns on cognitive labor past some point is very low. So unless you have more data or more compute. And he just doesn't think that. So I think that's the key disagreement. It's very hard to settle because it's very hard to run an experiment to determine who is right here. Because you would have to have some group of people who have tons of labor but they have like almost no compute and they're cut off from the rest of the world and like we can't run we can't run that experiment because it's the best you can do is if you have a team inside a company had less compute then how much does that affect their progress but in that case they're still benefiting from spillovers from other teams so that experiment is very unclean it's not a good proxy for uh what if the entire world has less compute and i think the, like my argument for thinking this is true, is that in a lot of the diverse domains, you have a tight correlation between by how much compute in that domain was being scaled, and by how fast

[18:26](https://realalpha.app/v/00iQmHkfYx8/?t=1106) you're making software progress in that domain. So for example, we've looked at this for a bunch of fields, we looked at for like chess, we looked at for SAT solvers, we looked at it for various different domains in AI, including natural progress, natural language processing, and computer vision and so on. This was where we were still at Epoch. And the finding is that the fastest software progress basically in any domain we have seen has been happening in AI. But the fastest compute scaling also in basically any domain we know is happening in AI. And the factor by which the AI thing is faster seems to be the same as a factor by which AI has been scaling inputs faster, scaling compute faster. And these numbers always appear to be around the same order of magnitude, which is kind of like a strange coincidence, if you think they should be independent.

[19:12](https://realalpha.app/v/00iQmHkfYx8/?t=1152) So, like, I think that's suggestive evidence. It's not decisive because it's not a causal experiment. It's just observational. But it's strong suggestive evidence that compute should, like, play a big role here. And I think they just think that, yeah, it plays a big role, but you can just have really clever researchers who devise really good experiments at small scales. They have incredible research tastes. And you can just, like, get a lot out of that. that you can use the existing compute 1,000, 10,000 times or whatever more efficiently than we're using it now. And that's the thing that doesn't seem plausible to me.

[19:44](https://realalpha.app/v/00iQmHkfYx8/?t=1184) We'll continue monitoring right after this message from our sponsors. 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 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. ClickHouse is the real-time database built to keep up with every query your apps and agents send its way. And it now runs Postgres, observability, and AI tooling on one platform. Learn more at clickhouse.com.

[21:05](https://realalpha.app/v/00iQmHkfYx8/?t=1265) So what do you expect the world to look like in 2029 such that you can say, I, Ege, was right and Daniel Kogutazer was wrong? I would want to know that because it's not clear to me. So today, Matthew, who was one of my co-founders, tweeted out asking people, so what is the thing we have to see in the real world for the people claiming there will be an intelligence explosion in 2027 or 2028 for their views to have been considered falsified? And I don't think if you can look at like AI 2027, you can look at the entire year of 2027. There's almost no like physical real world thing that happens in the entire 207, which is predicted, you know, in the story as a consequence of the internal explosion. The things that are predicted are these abstract metrics, like uplift, like compute multipliers, like whatever. But they're not like something you can actually see in the physical world. And so I am quite confused. Like, I personally don't know what these people are even predicting about a physical world to me it seems like they're saying look basically nothing happens in the physical world as a result of this until the moments we're all dead and of course if that's your story then of course like we won't be able to settle anything before all of us are dead

[22:27](https://realalpha.app/v/00iQmHkfYx8/?t=1347) but you can always say oh it's you know it's going to happen in the future or something and i mean it's maybe slightly uncharitable but like the issue is that i really don't understand how these people like could be considered to have been proven wrong i think they're not most of them are not making any concrete tangible prediction yeah so so i i also don't know like i don't know what observation would discriminate between what we think and what i think and what they think because i don't understand what predictions they're making sure yeah yeah um i guess i think that the the the prediction of like what will happen before we're all dead that would vindicate the ai 2027 view is like by 2029 or whatever by 2030 there are like robots all over the place doing physical world tasks yeah i think i think there's a decent chance of that happening though so i don't think by 2029 uh i mean there are already robots out there in the world doing physical tasks that already exists yeah yeah like um in industrial use cases and then maybe like unitary robots dancing in china or korea i don't know like the people have like cleaner like vacuum robots in their house yeah yeah vacuum that's like you have to actually

[23:38](https://realalpha.app/v/00iQmHkfYx8/?t=1418) define what they're talking about yeah yeah yeah i agree a lot of this stuff is very poorly defined and it's just difficult to define it seems it seems like kind of a fundamental problem more like um different criteria for bio improvements maybe that's something that's like somewhat physical um then it would be easier to like quantify uplifting capabilities but yeah yeah but i don't When I have to do this, right? Because like, I can try to figure out like, what do I predict? What do I think is likely going to happen? But it's I don't want to do that for their views. Because because I think they should be doing it for their views. Like I think me doing it on behalf of them could easily be misrepresenting what they actually think or whatever. So then they don't have to admit that they were wrong if things don't happen the way I interpreted their predictions. And just say, Oh, that's not actually what we meant or whatever. So I think they are the ones who should say.

[24:30](https://realalpha.app/v/00iQmHkfYx8/?t=1470) like here is what we expect to happen. And it's like, it's not always something that they don't, they don't do. Because like, in fact, Daniel has done this with some things before in the past. Like, for example, he said that we have this bet or he said, in three years, I will be able to, you know, describe some kind of ML experiments to whatever the best model is available at that time. And we'll just go ahead and implement it. And I won't have to like, deal with like the technical details like it will handle technical details i would just describe what the experiment i want to run is and it's just going to be able to run it and i said that's probably not going to happen i said maybe it's going to not happen in maybe like 50 chance that's in like 10 years or something and they thought it would happen in three years and i think that prediction was proven correct so it did happen in roughly three years um so that is the kind of thing that i would be looking for where like it's very easy to decide if it happened or not um but for the intelligent

[25:26](https://realalpha.app/v/00iQmHkfYx8/?t=1526) **Ege Erdil:** of exposure to be something that we should care about it has to be something with like real world sort of physical consequences so that's why that's the kind of prediction i would be looking

[25:34](https://realalpha.app/v/00iQmHkfYx8/?t=1534) for instead of predictions of products on abstract metrics which i think are less um less meaningful right so um more concretely uh the thing that caused a lot of ai safety people to update in the last week was this meter report on ai r&d automation and anthropic and it was two different things. One is like the plot of the internal Anthropic ECI over time inflecting upwards slightly, potentially due to AI acceleration. Meter believes it's mostly due to AI acceleration. And then the other thing is Meter says that they had a lot more evidence that Anthropic did not allow them to disclose. And so these two things cause the AI safety community to update, I think, more strongly and more quickly than I have ever seen them update before.

[26:22](https://realalpha.app/v/00iQmHkfYx8/?t=1582) Yeah. So, like, why do you not think this is an update? I mean, I think objectively, evidence looks kind of weak. Like, I think it's just this made-up metric that someone made up, on which we don't even know what it means to make faster progress, really. And we have some evidence that progress has sped up, but we don't really understand the y-axis of that metric very well. And there's a bunch of redacted evidence by a company which has a track record of being paranoid about this topic.

[26:52](https://realalpha.app/v/00iQmHkfYx8/?t=1612) **Ege Erdil:** So I just see it as very weak evidence.

[26:55](https://realalpha.app/v/00iQmHkfYx8/?t=1615) And I think what's happening with the people updating way too much is some kind of social preference dynamic where the influential people update, and therefore other people feel like if they don't update, then they are out of step with the crowd and so on. I think often people's updates about these questions are very just socially correlated. And people care way more about what do the people I respect think, I'm like, am I going to be, am I going to be appearing like out of step, unreasonable if I just don't say, look, I don't, I just don't find this that convincing. I think that's a lot of what's going on.

[27:30](https://realalpha.app/v/00iQmHkfYx8/?t=1650) And it is happening in the other direction as well. Like, I don't know, like people updated to be more bearish end of 2025 and now they updated to be way more bullish. When I just don't see anything that is like that informative, personally. Like, I haven't updated very much since end of 2025 at all. Like, I don't think we've seen anything that is like very decisive evidence. in like either direction but i don't know so i don't i don't really understand what's going on People make a very big deal out of something like this. Do you think full automation of white-collar work is still 30 years out?

[28:04](https://realalpha.app/v/00iQmHkfYx8/?t=1684) 30 years out from now. I mean, I said it was 30 years out some time ago. So, you know, it's— Like a couple years ago. It's still roughly 30 years out? I think full automation of white-collar work maybe was like 20 years out. Okay. All right. Well, final question before we go. After you, we're going to talk to Baksh Ligaris, who's the CEO of Redwood Research and is, you know, much more of a doomer and has also updated towards, you know, know much shorter timelines so like what should we ask him um i guess you can just ask him for the for the thing that we were discussing right we're like uh what would you have to see over the next two or three years for you to admit that you're wrong and i think like like one thing that i can say is is that um i don't know like like if we if we have if like like global primary consumption option goes up by 10x over the next three years like it's like a very hard metric where i was like yeah i was wrong you know like and uh i want to understand if they're making that prediction because i think they might say no no we actually don't predict that like we don't think that's going to happen but then i want to understand what is the thing they do predict is going to happen

[29:11](https://realalpha.app/v/00iQmHkfYx8/?t=1751) so i can figure out like do i actually disagree with that because like sometimes it's actually not even clear like what we're talking about yeah right well this is super interesting um Yeah, we love talking to, like, informed short timeline skeptics because there's so few of them left. So, yeah, keep fighting the good fight. Thanks so much, Ege, for coming on MTS. Thanks, Ege. Bye. 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.

[30:21](https://realalpha.app/v/00iQmHkfYx8/?t=1821) visit getvcx.com for more info that's getvcx.com 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
