Language Models, Game Transcripts
Martin discusses the fascinating ability of language models to make legal moves in a game without any prior knowledge of the rules. He also explores the potential interpretability of the models' internals and how they can adapt to gender-specific language in different contexts. Understanding how these models represent users can have implications for trust, safety, and personalized responses.In this clip
From this podcast

The Gradient
Martin Wattenberg: ML Visualization and Interpretability
Related Questions