Latest / Minds, Bodies, and Terawatts / Ex-Meta Researcher Launches 'Recursive Superintelligence' AI Startup
Transcript
- 0:00[Host] This is Minds, Bodies, and Terawatts, where the future funds itself. A startup called Recursive Superintelligence just emerged from stealth with six hundred fifty million dollars and a four-point-six-five billion dollar valuation. Google Ventures and Greycroft led the round. Nvidia and AMD wrote checks too.
- 0:20[Guest] That name is doing a lot of work. Most AI companies hide behind vague mission statements. This one put recursive superintelligence on the letterhead — AI that redesigns itself in an accelerating loop. Eight co-founders, fewer than thirty employees, no product yet.
- 0:39[Host] Who is actually driving this? When the valuation is north of four billion before there is a product, the names on the cap table matter more than the pitch deck does.
- 0:53[Guest] Richard Socher, former chief scientist at Salesforce, runs it. But the co-founder drawing headlines is Yuandong Tian, a senior researcher at Meta's FAIR lab.
- 1:04[Host] What is his story? Because going from getting laid off to raising hundreds of millions in six months is not a normal career arc.
- 1:15[Guest] He was laid off last October when Meta cut six hundred people across its AI division. His post on X was pointed — the people who should really be responsible for solving the problems, he wrote, are not the ones being laid off. Six months later he is co-founding this.
- 1:38[Host] Meta's AI cuts are now the origin story for the most explicit superintelligence play on the market. Why did Nvidia and AMD write checks? They sell GPUs to everybody.
- 1:51[Guest] Because recursive self-improvement is the ultimate compute customer. A system that redesigns itself needs endless cycles of training, evaluation, retraining. For chip companies, this is investing in their own demand curve.
- 2:05[Host] But thirty people, no shipping product, and a valuation that dwarfs what OpenAI or Anthropic carried through their first several rounds. What is the market actually buying?
- 2:18[Guest] The roster. Peter Norvig co-authored the standard AI textbook. Jeff Clune pioneered AI-generating algorithms at DeepMind. Investors saw the specific team most likely to crack autonomous self-improvement. The round was heavily oversubscribed.
- 2:32[Host] There is a geopolitical wrinkle. Tian is Chinese-born. Washington spent two years tightening chip export controls to slow China's AI development, and now American chip money is backing a Chinese-born scientist on this project.
- 2:48[Guest] The tension is real but overstated. Tian has worked in American labs for over a decade. The company operates from San Francisco and London. Export controls target hardware shipments to China, not researchers inside the United States.
- 3:04[Host] Every major AI CEO staked a prediction on AGI arriving between 2027 and 2030. Does this company move the needle?
- 3:14[Guest] It changes the texture. Before this week, recursive self-improvement was a theoretical concern in alignment papers. Now it has a business plan, a cap table, and a shipping address. Whether they succeed is open. Whether the attempt changes what regulators need to prepare for is already answered.
- 3:35[Host] Picture thirty researchers in a San Francisco office building the machine designed to make every other office obsolete. That is what Patrick Déglon calls the Labor Cliff in Unscarcity — the point where AI automates faster than new work appears, and now someone is explicitly trying to accelerate it.
- 3:57[Guest] And the world has not decided who watches them try. The Unscarcity essay on the 2026 AGI Timeline lays out exactly what Altman, Hassabis, and Amodei have publicly predicted — and a venture-backed superintelligence shop tightens that window. For more on the race to self-improving AI and who gets caught in the transition, head to unscarcity dot ai for the articles and the forum. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.