Gonçalo Esteves / building capable systems Field Notes →

The Vault / People

Person

Demis Hassabis

Created 2026-11-14, promoted off the Moonshots 311 — Jensen's Departure, NVIDIA's succession plan, AI doomer response, and Demis talks RSI (Live) decomposition (owed since Moonshots 310 — Anthropic's Trillion-Dollar IPO, NVIDIA's Fallout, and Elon's 30M Robot Workforce).

status · researched added 14 Nov 2026 updated 14 Nov 2026 from Moonshots #311 · Google DeepMind personaiai-safetybiology see it in the graph →

Created 2026-11-14, promoted off the Moonshots #311 decomposition (owed since Moonshots #310).

Who he is (one line): co-founder and CEO of Google DeepMind; shared the 2024 Nobel Prize in Chemistry for AlphaFold (protein structure prediction); also runs Isomorphic Labs (AI drug discovery).

Why it is in this archive

Gate: named in 41 episodes in the regex index (the research archive, 2026-11-14) with no page. On Moonshots #311 he is a subject: the title ("Demis talks RSI") and a two-minute clip that frames an 18-minute segment. Elsewhere the panel uses him as a frame: the Nobel as proof that AI does science, AlphaFold's co-creator John Jumper moving to Anthropic, and Will Marshall's proposed "conclave" of AI leaders (Moonshots #266).

The position on tape (EP311)

  • Open- vs closed-loop RSI. Today's Recursive self-improvement (RSI) makes "a human engineer maybe three, four, five times more productive" with the human still reviewing; "the next phase which I think carries the most risk is closed RSI closed loop where you basically manage to somehow figure out how to do it without the human reviewer"; the result "will likely be quite alien"; "we should be very cautious", but there is "this sort of intense… competitive race to improve your coding systems" (clip, 0:58:03–0:59:54). Venue of the clip not stated on air; not checked.

What the panel said

episode who what ts
Moonshots #311 Alexander Wissner-Gross "everyone's doing it already including his lab Google very publicly is leaning into RSI"; "alien" is "fundamentally misguided" 1:04:10, 1:01:52
Moonshots #311 Emad Mostaque "How you going to do human in the loop for a 10,000 token a second model and there's 100,000 of them? You're never going to do it" 1:04:43
Moonshots #311 Salim Ismail takes "alien" as the good news: intelligence "orthogonal and complementary… not replicative" 1:00:36
Moonshots #311 Alexander Wissner-Gross on disease, Demis "could be more wholesomely articulating it the way Daario is", though Isomorphic is "walking the talk" 1:23:50–1:24:12

The pattern: the panel cites him as the scientist who proves AI can do science, then sets aside his caution as the lab leader. On EP311 every mate rejects the closed-loop warning in the same segment.

Do not say

  • That DeepMind runs closed-loop RSI. That is Alex's claim, and Demis says the opposite.

How this page stays current

  1. Grep new transcripts for demis|hassabis|assabis|abis. Diff: any new position on RSI, AGI timelines or disease.
  2. DeepMind / Isomorphic announcements. Diff: a released result the panel cites.

Related

Google DeepMind · Recursive self-improvement (RSI) · Dario Amodei · Paul Christiano · Moonshots #311

Reading rule. Anything marked unverified or on-air is a claim made on the podcast and has not been checked against a primary source yet. It stays marked until it has. Sourced numbers link to where they came from.