ICALens: Interpretability Method for Language Models Without Training Additional Autoencoders11. June 20264. July 2026AI ModelsICA-based analysis enables rapid exploration of interpretable directions in language models without expensive training of additional autoencoders. Share on:
