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.
Why do you think that https://github.com/CSSLab/maia-chess is a good alternative to chess_llm_interpretability