# Thore Graepel argues in MIT Technology Review that LLMs don’t reason

[Canonical reading](https://escapemeridian.com/readings/thore-graepel-argues-in-mit-technology-review-that-llms-don-t-a70c5d39)

MIT Technology Review published this newsletter on October 2, 2026. In its opinion teaser, Thore Graepel argues that today’s AI lacks the reasoning powers he attributes to AlphaGo, the program he helped build. He recalls that AlphaGo beat Go champion Lee Sedol after making a move so unusual that some commentators thought it was a programming glitch. Graepel says that choice showed creative reasoning, and argues that AI needs a fresh approach to machine reasoning drawing on AlphaGo’s architecture. He says this belief led him to leave his position at Google DeepMind. The teaser introduces his case but does not explain what aspects of AlphaGo’s architecture he thinks could support that approach.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

## Original source

MIT Technology Review · Article

Original source date: Oct 2, 2026 (2026-10-02T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-20T03:18:09.730Z. First observed: 2026-11-20T02:22:03.941Z. Retrieved: 2026-11-20T02:24:04.253Z.

- [The Download: a biological de-aging contest and why LLMs don’t reason](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Limits

Text as retrieved on November 20, 2026; originally published October 2, 2026. The newsletter provides a short teaser for Graepel’s opinion, not the full argument or details of the proposed approach.


## Open question

What aspects of AlphaGo’s architecture does Graepel think could support the fresh approach to machine reasoning he calls for?


## Source-stated

Graepel argues that today’s AI lacks the reasoning powers he attributes to AlphaGo.

> It was AlphaGo’s powers of reasoning that made this creative choice—and these are powers that today’s AI lacks.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source-stated

Graepel says his belief that AI needs a fresh approach to machine reasoning drawing on AlphaGo’s architecture led him to leave Google DeepMind.

> This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning, one that draws on AlphaGo’s architecture.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source-stated

Graepel recalls that AlphaGo beat Lee Sedol with a move some commentators thought was a programming glitch.

> AlphaGo won after making a move so strange that some commentators thought it was a programming glitch.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.


## Source media

Media retrieval status: unchecked.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## Large language model

A language model with many learned numerical settings, used to process and generate language.

- [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## Editorial provenance

Version 1. Publication ID: a008a380-4ed5-144f-5ee8-6a805e3031d5.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T03:14:05.754Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 4a86ec17cb0c521c9e02943131d8305b800f476b842f14da6ee8a3f8c7aabe60. Source text hash: 4196f018349688aef041dd997141ecfd2e70ddac99ae31fecd94082b4380d415.


## Revision and source-check history

### Version 1 — 2026-11-20T03:18:09.730Z

MIT Technology Review published this newsletter on October 2, 2026. In its opinion teaser, Thore Graepel argues that today’s AI lacks the reasoning powers he attributes to AlphaGo, the program he helped build. He recalls that AlphaGo beat Go champion Lee Sedol after making a move so unusual that some commentators thought it was a programming glitch. Graepel says that choice showed creative reasoning, and argues that AI needs a fresh approach to machine reasoning drawing on AlphaGo’s architecture. He says this belief led him to leave his position at Google DeepMind. The teaser introduces his case but does not explain what aspects of AlphaGo’s architecture he thinks could support that approach.

## Original source

MIT Technology Review · Article

Original source date: Oct 2, 2026 (2026-10-02T12:10:00Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-20T03:18:09.730Z. First observed: 2026-11-20T02:22:03.941Z. Retrieved: 2026-11-20T02:24:04.253Z.

- [The Download: a biological de-aging contest and why LLMs don’t reason](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Limits

Text as retrieved on November 20, 2026; originally published October 2, 2026. The newsletter provides a short teaser for Graepel’s opinion, not the full argument or details of the proposed approach.


## Open question

What aspects of AlphaGo’s architecture does Graepel think could support the fresh approach to machine reasoning he calls for?


## Source-stated

Graepel argues that today’s AI lacks the reasoning powers he attributes to AlphaGo.

> It was AlphaGo’s powers of reasoning that made this creative choice—and these are powers that today’s AI lacks.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source-stated

Graepel says his belief that AI needs a fresh approach to machine reasoning drawing on AlphaGo’s architecture led him to leave Google DeepMind.

> This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning, one that draws on AlphaGo’s architecture.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source-stated

Graepel recalls that AlphaGo beat Lee Sedol with a move some commentators thought was a programming glitch.

> AlphaGo won after making a move so strange that some commentators thought it was a programming glitch.

- [Supporting source](https://www.technologyreview.com/2026/10/02/1145666/the-download-biological-de-aging-ai-reasoning/)

## Source and contributor context

MIT Technology Review · Publisher

The publisher is identified here. No individual author has been inferred.


## Source media

Media retrieval status: unchecked.


## Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.


## Large language model

A language model with many learned numerical settings, used to process and generate language.

- [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## Editorial provenance

Version 1. Publication ID: a008a380-4ed5-144f-5ee8-6a805e3031d5.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T03:14:05.754Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical.

Observation: 4a86ec17cb0c521c9e02943131d8305b800f476b842f14da6ee8a3f8c7aabe60. Source text hash: 4196f018349688aef041dd997141ecfd2e70ddac99ae31fecd94082b4380d415.


### passed — 2026-11-20T03:14:05.754Z — anthropic/claude-opus-5.5

Publication: a008a380-4ed5-144f-5ee8-6a805e3031d5

attribution — pass: The reading attributes the argument to Thore Graepel, in an opinion teaser in MIT Technology Review's The Download newsletter. The newsletter comments on items in its own words, so it counts as one piece. The reading stays within the Graepel item and keeps its claims attributed to him.

> Opinion: Don’t be fooled—LLMs don’t reason
—Thore Graepel

summary — pass: The title, summary and question match the teaser. The reading identifies Graepel, AlphaGo and the Lee Sedol match, and says it is a short teaser rather than the full argument. LLM is defined in the page definitions. The question names a real gap the teaser leaves open: which parts of AlphaGo's architecture he would draw on. The reading keeps the force of his claim, including his departure from DeepMind.

> Here’s why today’s AI doesn’t really reason—and what it would take to change that.

date — pass: The reading gives October 2, 2026, the UTC calendar day of originalPublished. That matches the URL path. modifiedAt (11:48 UTC) falls before the publication instant. The reading uses no hindsight and no later must-read items. The source's own 'ten years ago' is not repeated. The evidence is a publisher page, so no archive disclosure is needed.

> This is today's edition of The Download, our weekday newsletter

support — pass: Each claim matches the quoted source text: AI lacks AlphaGo's reasoning powers, the glitch-like move, and leaving DeepMind because he believes a fresh approach drawing on AlphaGo's architecture is needed. The reading adds no unsupported benefits, costs or forecasts.

> This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning, one that draws on AlphaGo’s architecture.

media — pass: No media is selected and no verified candidates were supplied, so no media is the appropriate choice.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/a70c5d3917448fb408dd15feffed2027782a39287209f4db6bb217ba927c1a03/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/thore-graepel-argues-in-mit-technology-review-that-llms-don-t-a70c5d39.json)
