adamkarvonen/chess_llm_interpretability

Visualizing the internal board state of a GPT trained on chess PGN strings, and performing interventions on its internal board state and representation of player Elo.

121 best points · 2 comments · 1 Hacker News post · 221 GitHub stars · View on GitHub

Hacker News discussions

• Chess-GPT's Internal World Model (121 points, 2 comments, by anjneymidha, 2024-01-06)