Google DeepMind
The AI laboratory where alpha-go was built and where thore-graepel worked from 2015 to 2021 and again from 2025 to 2026. The lab matters to this vault in two ways.
AlphaGo came out of it, and that program is the central example in graepel-2026-llms-dont-reason: intuition from learned networks, deliberation from explicit search over a game tree, the combination producing move 37.
The vault’s model-collapse line also runs near it. ilia-shumailov is the first author of shumailov-2024-model-collapse and a co-author of the reasoning-trace extraction paper cited in chain-of-thought-faithfulness. The Nature paper’s affiliation detail is not in the vault’s record of it, so no claim about DeepMind’s role in that work goes here beyond what those pages support.
Graepel’s own site dates his second stint, 2025 to 2026, to a Post-AGI team, and his article says the departure was over the position he defends in it. Wikipedia describes the role as research group lead working on AGI futures.
Sources
- graepel-2026-llms-dont-reason (departure statement, byline, read 2026-10-04)
- https://thoregraepel.github.io/ (stint dates, Post-AGI team, read 2026-10-04)
- https://en.wikipedia.org/wiki/Thore_Graepel (research group lead description, read 2026-10-04)