Chris Potts Pushes Back on Wilson's Interpretability Critique: Focus on Goals, Not Methods
ShahabBakht · x · 2026-11-02
Stanford professor Chris Potts responded to Andrew Wilson's critique of the interpretability field, arguing he conflated the whole field with a culturally prominent subset of its work.
Potts's core point: criticizing individual methods is healthy, but the underlying goals matter most. Influence functions may be the wrong method, yet solving data attribution would be transformative; saliency maps are clearly inadequate, but the broader family of feature attribution methods is far better. He also asked whether critics had checked whether researchers had already branched out beyond the methods being criticized.
Related event: Stanford's Potts Rebuts Skeptics, Defends Mechanistic Interpretability(2 posts)→
More from AGI Musings
- AI is eating cost centers: mid-career advice says move to profit centers now — emeka_boris · 2026-11-02
- Pinboard founder: alignment debates echo 1947 atomic scientists' flawed binaries — nitarshan · 2026-11-02
- Eric Topol asks whether medicine will face its own 'bitter lesson' from AI — EricTopol · 2026-11-02
- Measuring AI agents wrong: task completion counts less than the human verification work left behind — alifcoder · 2026-11-02
- Kyle Kingsbury: LLMs erode trust in software and its documentation — Simon Willison · 2026-11-02
- NYT essay: AI consciousness is 'definitely possible' — we don't even understand consciousness — coherence · 2026-11-02