Using AlphaGo to Stop AI Hallucinations

Thore Graepel used to work at DeepMind. Now he runs a startup called Metis Reasoning. He thinks he can stop AI models from making things up by using the tech behind AlphaGo.

AI usually hides knowledge deep inside its system. Graepel wants to build a clear record of facts instead. This would let people trace every answer back to real evidence.

The idea sounds promising for making AI more trustworthy. However, the startup has not built a working product or shared test results yet.