Deepmind researcher proposes AlphaGo-style reasoning to cut AI hallucinations
Thore Graepel, a former Deepmind researcher, argues that AI hallucinations could be reduced by adapting AlphaGo’s structured reasoning approach. His startup, Metis Reasoning, proposes separating knowledge from model weights and storing it in an external ledger. A verification module would then check assumptions against evidence, re-evaluating conclusions if assumptions change—similar to how…
Key points
- Metis Reasoning proposes an external knowledge ledger to verify AI assumptions, reducing hallucinations and enabling explainable outputs
- Graepel’s approach separates knowledge from model weights, inspired by AlphaGo’s structured search for valid moves
- Metis has no working system yet, but Graepel is raising seed funding and targeting edge and robotics applications
Graepel, who co-developed AlphaGo and worked on Deepmind’s post-AGI team, presented the idea at Nvidia’s GTC conference. His team acknowledges challenges like partial real-world data and the cost of evidence but claims the approach could address AI’s lack of economic impact in fields like medicine or robotics. Metis has not yet built a working system or published benchmarks. Graepel is reportedly raising a low tens-of-millions seed round, with plans for a larger follow-on investment.
Former Deepmind researcher thinks AlphaGo's playbook can cure AI hallucinations
The Decoder · 27 October 2026
Loading the full article…
This text was published by The Decoder and written by Jonathan Kemper. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Unite.AI breaks down how large language models learn from data · 2 src
- OpenAI model allegedly disproves Exponential Time Hypothesis · 7 src
- Researchers combine GPT-Neo with Random Forest to predict Agrobacterium promoters at 97% accuracy · 1 src
- Coalition study shows non‑English languages use 1.5‑5x more AI tokens · 1 src
- IBM and Kaggle launch ScarfBench migration leaderboard for coding agents · 1 src
Comments
via GitHub Discussions