List your company

Feature Interpretability on LLM Activations

86.9AutoInterp ScoreHSAE
80.45282.12683.885.474Feb 12, 2026

Evaluation results

MethodLinks
HSAEDictionary Size=163842026.0286.932.4
BaselineDictionary Size=163842026.0286.133.8
HSAEDictionary Size=81922026.0285.931.4
BaselineDictionary Size=81922026.0285.432
BaselineDictionary Size=40962026.0282.930.2
HSAEDictionary Size=40962026.0281.428.6
BaselineDictionary Size=20482026.0281.324.5
HSAEDictionary Size=20482026.0280.727.2