Numerai founder Richard Craib: Situational Awareness's mistake was volatility, not leverage
Key Points
- Numerai founder Richard Craib argues Situational Awareness's collapse stemmed from running 150% volatility, not leverage itself; he runs more leverage at one-tenth the volatility, putting the funds in different risk regimes entirely.
- Craib dismisses Situational Awareness's 400% return in six months as mathematically inseparable from a coin flip, contrasting it with Buffett's 5,000,000% gain built on thirty years of survivable strategy.
- Going long AI stocks generates a risk premium any investor can replicate directly, not alpha; true alpha requires hedging out market beta and thousands of other factors to produce uncorrelated returns.
Summary
Read full transcript →Volatility was the mistake, not the leverage
Richard Craib runs Numerai, an AI-driven quant hedge fund he founded in San Francisco in December 2015. The fund trades equities using machine learning, targets true market-neutral alpha, and has grown AUM roughly 100% per year for the past two years. His core argument about Situational Awareness's blowup cuts against the popular narrative — the problem wasn't that Leopold Aschenbrenner used leverage. It's that he ran catastrophic volatility.
Risk is volatility squared. Craib says Numerai operates with more leverage than Situational Awareness did, but at roughly one-tenth the volatility. At 150% volatility, Situational Awareness was carrying 150-squared units of risk. At roughly 10% volatility, Numerai is in a categorically different regime. Being up 400% in six months, Craib argues, is mathematically inseparable from being a coin flip — you cannot generate that return without the volatility that makes a full wipeout equally plausible.
“I was mostly critical about the high volatility. I think that was the mistake, not the leverage. If they have 150 squared units of risk and we have 10% squared, it is a whole different ballgame... All you need to do to be Warren Buffett is just get 9% alpha per year and add that to the S&P. And all the people taking too much risk are going to lose it all over and over and over again.”
Alpha versus risk premium
Craib's sharper critique is definitional. Generating a return by going long AI stocks is not alpha — it's collecting a risk premium that any LP can replicate themselves. There is nothing stopping a sophisticated investor from buying Anthropic shares or going long Micron directly, without paying fund fees. Any fund that packages that trade and calls it alpha is, in Craib's framing, "almost childish" about what it's selling.
True alpha, by his definition, is return that survives after hedging out every factor: market beta, momentum, sector, size, and thousands of others. It's uncorrelated with everything. Citadel and Millennium are trying to manufacture the same thing.
The Warren Buffett comparison
Aschenbrenner was up roughly 400% in the first half of 2026. Craib's response is blunt: Buffett is up approximately 5,000,000% by being sensible for thirty years. The choice isn't between courage and cowardice — it's between a strategy that survives long enough to compound and one that almost certainly doesn't. The mathematically correct amount of leverage for a given Sharpe ratio is smaller than most people think, and the big institutional funds running tight vol targets are already operating near that optimum, not below it.
Craib adds that a long investment horizon is a prerequisite, not a preference. If you believe the singularity arrives in 2030, the market doesn't — volatility would look completely different if it did — and building a portfolio around that private timeline is a bet the market is explicitly not pricing.
Where he gives Situational Awareness credit
Craib says there is a legitimate market for high-conviction, high-risk vehicles. He was an early LP in Polychain, the crypto fund he describes as the crypto-market equivalent of Situational Awareness, and says the funds are "cool" — many institutional investors have a marginal allocation they explicitly want to run hot, because their core portfolio is already diversified across bonds and equities. Losing 500 million in that tranche might still leave the overall portfolio positive on the day. He says he doesn't fault the LPs.
His closing critique is cultural rather than financial: stop treating strong recent performance as evidence of genius. AI is already embedded in the market itself — in Citadel's systems, in Numerai's models — so a thesis that amounts to "buy AI stocks because AI will be big" is betting against a market that has already processed that view.
Takeaway: Craib's case is that Situational Awareness ran a vol level that made a large drawdown close to inevitable, regardless of the underlying thesis quality. The rescue via Citadel means Aschenbrenner may yet get a second act, but the lesson Craib draws is mechanical — run survivable volatility or accept that compounding is impossible.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.