# Ex-DeepMind Researcher Thore Graepel Launches AI Lab Metis Reasoning Metis Reasoning aims to replace LLM scaling with explicit, verifiable search-based reasoning architectures. - **Source:** [The Next Web](https://brifing.si/source/afe24b20-b275-4c19-b9a6-8bb3c8babae7) - **Published:** 2026-10-20 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/e2340875-3dec-452d-8e64-ea5058eb6d1d ## Founding Vision and Leadership Thore Graepel, a key figure behind Google DeepMind’s AlphaGo, has launched Metis Reasoning in London. He is joined by co-founders Piers Ingram, Michael Warren, and David Berman. The team currently consists of 12 professionals operating across offices in London, Singapore, and Los Angeles, focusing on fundamental shifts in artificial intelligence architecture. ## The Core Reasoning Problem Metis Reasoning argues that current large language models cannot be trusted for high-stakes tasks in engineering, science, medicine, or law. Their white paper, The Age of Reasoning, highlights that LLMs hallucinate and lack transparency. Because reasoning is buried within model weights, users cannot audit the knowledge behind specific generated answers. ## Lessons from AlphaGo Move 37 The founders draw inspiration from AlphaGo’s famous Move 37 against Lee Sedol in 2016. While experts initially dismissed the move as a blunder, it was actually a product of search and evaluation. Metis aims to replicate this capability, moving beyond simple prediction to a system that prioritizes explicit, verifiable reasoning paths. ## Mechanics of the Reasoning Engine The proposed system functions as a game between an inquirer and nature, based on philosopher Jaakko Hintikka’s theories. It maintains an explicit record of known information and confidence levels. A separate judgment system accepts claims only when supported by evidence, selecting subsequent steps based on their ability to reduce overall uncertainty. ## Efficiency and Model Scalability Metis rejects the industry trend of ever-larger models and ballooning training budgets. By utilizing search-based reasoning, the company claims its approach is significantly more compute-efficient than current frontier models. This design allows for smaller models that can perform complex tasks, potentially enabling reasoning capabilities on edge devices and robotics platforms. ## Knowledge Management and Updates A key advantage of the Metis architecture is the ability to update or remove specific knowledge without requiring full model retraining. Because the system tracks reasoning steps explicitly, users can inspect and verify the logic at every stage. This modularity offers a distinct alternative to the opaque nature of standard transformer-based architectures. ## Current Status and Funding While the company has published its white paper, it has not yet released a commercial product or specific test results. Co-founder Piers Ingram noted that early internal results on frontier reasoning problems are encouraging. Although reports suggest the firm is raising tens of millions, Ingram stated funding is currently more than adequate. ## London AI Ecosystem Context Metis Reasoning joins a growing cluster of London-based AI labs founded by DeepMind alumni, including David Silver’s Ineffable Intelligence. This trend highlights a shift toward highly specialized, philosophical approaches to AI development. These startups aim to solve the inherent limitations of scaling by focusing on structural innovations rather than just raw data. ## Sources covering this story - https://thenextweb.com/news/metis-reasoning-thore-graepel-alphago-london-ai-lab — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._