Bear Häon

Background * The AI safety field has spent a decade building tools for systems trained to think and digitally act. The next decade will likely deliver widely deployed robot foundation models (RFMs): systems trained to think and physically act. * A year ago, we published the first mechanistic interpretability paper...