Thore Graepel, a former researcher at DeepMind, is advocating for a novel approach to mitigate AI hallucinations by leveraging the search principles used in AlphaGo. His startup, Metis Reasoning, proposes the use of an explicit knowledge ledger that would allow AI conclusions to be verified against evidence, rather than relying solely on model weights.
Graepel’s vision contrasts with traditional AI models that often lack transparency in their reasoning processes. By making every conclusion checkable, this approach aims to enhance the reliability and trustworthiness of AI systems. However, as of now, there is no functioning system or established benchmarks to validate this concept.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
