Former Deepmind researcher's plan to cut AI hallucinations with AlphaGo's playbook

thore graepel, a former deepmind researcher, founded metis reasoning to develop ai systems that track what they know and verify conclusions against evidence. the approach uses a knowledge ledger to separate knowledge from model weights, similar to alphago's architecture, and an independent module checks assumptions against calculations, formal proofs, or empirical evidence. this aims to reduce hallucinations and provide source-backed explanations. metis reasoning's method could favor smaller models running on edge devices and in robots, potentially changing how companies deploy ai. the approach currently exists as a theoretical framework with no working system.