Thore Graepel, a former researcher at Google DeepMind known for work related to AlphaGo, has launched a new AI startup in London called Metis Reasoning. The company focuses on the claim that large language models are not sufficiently reliable for reasoning tasks on their own.
Metis Reasoning says it aims to use ideas associated with AlphaGo—where progress comes from structured approaches to decision-making and learning—to broader, general reasoning capabilities. The startup’s approach is presented in a white paper, “The Age of …” (as cited by one outlet), setting out the rationale behind its thesis and how it intends to build systems that reason more robustly.
Across the available coverage, both outlets frame the announcement similarly: a founder with DeepMind ties launches a London-based AI lab/startup, and the central differentiator is skepticism about current language-model reasoning reliability, alongside a commitment to apply lessons from AlphaGo to wider reasoning problems.