Stanford's Chris Potts critiques auto-interpretability evaluation and clarifies stronger models are easier to interpret

Stanford professor Chris Potts criticized auto-interpretability evaluation for neglecting precision and recall and relying on heuristic spot-checks that invite 'just-so stories,' while clarifying that stronger models are actually easier to interpret and agentic tools are aiding interpretability research.

2026-10-19 ~ 2026-10-19 · 2 related posts

• Chris Potts critiques auto-interp evaluation: precision, recall, and the case for ex ante testing — ChrisGPotts · 2026-10-19
• Chris Potts clarifies: better models are easier to interpret, and agentic tools speed up interp research — ChrisGPotts · 2026-10-19