Peter Diamandis
Episode overview

Recursive's $670M Bet on Self-Improving AI, Sonnet 5.5 Hits 70%, Elon Co-Leads Pentagon Push EP 299

Peter Diamandis · 3 October 2026 · 2h 25m · 3 Egleze moments
Recursive's $670M Bet on Self-Improving AI, Sonnet 5.5 Hits 70%, Elon Co-Leads Pentagon Push EP 299 ▶

What happened in this episode?

Peter Diamandis hosts Richard Socher, chief executive of Recursive and founder of You.com, for a wide-ranging discussion on AI-driven scientific discovery and the prospects for recursive self-improvement (RSI). Diamandis says Recursive has raised $670m from named investors and secured $410m in AWS compute, while disclosing he is a seed investor. Socher argues scientific progress has slowed due to fragmentation across disciplines and says “full stack” AI — combining large language models, digitised scientific data, simulation, robotics and agent “swarms” — can compress the loop between hypothesis, experiment and validation.

The episode’s most consequential segment centres on RSI and artificial superintelligence (ASI). Socher says AI’s new ability to write code has made “weak forms” of RSI possible already, but predicts several decades before AI surpasses “all of humanity combined” across multiple dimensions of intelligence. He disputes more imminent ASI timelines attributed to Elon Musk, while acknowledging near-term superhuman performance in narrower domains.

On policy and safety, Socher criticises proposals he says would ban RSI, arguing enforcement would require a “totalitarian surveillance state”. He also calls Anthropic’s constitutional AI framing “fake”, alleging the lab’s cyber-security work contradicts its stated “hard constraints”. Elsewhere, he claims Parallel Bio obtained FDA approval to skip animal trials using organoids, and warns that compute scarcity remains a real bottleneck, citing rising prices for high-end GPU clusters and older H100 chips.

What are the key points?

Who is speaking?

Named speakers identified in Egleze's source-linked moments from this episode.

Which topics are covered?

How is this page sourced?

Programme
Peter Diamandis
Published
3 October 2026
Duration
2h 25m
Original episode
Open the original source ↗

Egleze verifies the speaker, source and timestamp for each linked moment. Unless a moment is explicitly marked independently verified, statements remain attributed claims by the named speaker rather than established facts.

Watch original episode More from Peter Diamandis

3 moments from this episode

Source-linked · editorially selected
01
AI & Tech

Host Peter Diamandis says Recursive raised $670m and received $410m in AWS compute

Peter (likely Peter Diamandis) · at 3:23 · Attributed speaker claim
Richard, congrats on a $670 million fund raise from Google Ventures. Graycraft, Nvidia, and AMD and 410 million in compute from AWS.

In the episode introduction, host Peter Diamandis congratulates guest Richard Socher and says Socher’s company, Recursive, has raised “$670 million” (naming Google Ventures, Graycraft, Nvidia and AMD) and received “$410 million in compute from AWS”. Diamandis adds that he is a seed investor in Recursive.

02
AI & Tech

Socher says Parallel Bio won FDA clearance to skip animal tests

Richard Socher · at 17:17 · Attributed speaker claim
They build tiny organoids on a petri dish, right? And get this, if you love animals, you too can love AI. Why? Because they got FDA approval to skip animal trials. Turns out we cured most diseases in mice. It's not that helpful.

Richard Socher says organoid-testing firm Parallel Bio has received FDA approval to skip animal trials, a claim that remains unverified. He argues lab-grown organoids can predict human drug responses and toxicity better than mouse models.

03
AI & Tech

Richard Socher argues AI self-improvement bans would require mass surveillance

Richard Socher · at 1:21:04 · Attributed speaker claim
I would argue and this is maybe a strong stance to really truly enforce no recursive self-improvement for instance you would need a totalitarian surveillance state the likes of which humanity has never seen because anyone can have a GPU on their little laptop and ask that AI to just like improve its harness.

Richard Socher argues that banning recursive self-improvement in AI would be impractical to enforce without pervasive monitoring of private use. He says people could run small self-improvement loops on local hardware, and that serious enforcement would resemble “thought police” oversight.