flâneur — a map of the web's best reading

Causal Scrubbing: a method for rigorously testing interpretability hypotheses [Redwood Research] — LessWrong

lesswrong.com · 11,802 words · saved by 4 readers

Summary: This post introduces causal scrubbing, a principled approach for evaluating the quality of mechanistic interpretations. The key idea behind causal scrubbing is to test interpretability hypotheses via behavior-preserving resampling ablations. We apply this method to develop a refined understanding of how a small language model implements induction and how an algorithmic model correctly classifies if a sequence of parentheses is balanced. A question that all mechanistic interpretability work must answer is, “how well does this interpretation explain the phenomenon being studied?”. In the many recent papers in mechanistic interpretability, researchers have generally relied on ad-hoc methods to evaluate the quality of interpretations.[1] This ad hoc nature of existing evaluation methods poses a serious challenge for scaling up mechanistic interpretability. Currently, to evaluate the quality of a particular research result, we need to deeply understand both the interpretation and t

x Causal Scrubbing: a method for rigorously testing interpretability hypotheses [Redwood Research] — LessWrong [Redwood Research] Causal Scrubbing Redwood Research Causal Scrubbing Interpretability (ML & AI) AI Frontpage 208 Causal Scrubbing: a method for rigorously testing interpretability hypotheses [Redwood Research] by LawrenceC , Adrià Garriga-alonso , Nicholas Goldowsky-Dill , ryan_greenblatt , jenny , Ansh Radhakrishnan , Buck , Nate Thomas 3rd Dec 2022 AI Alignment Forum 24 min read 35 208 Ω 103 * Authors sorted alphabetically. Summary: This post introduces causal scrubbing, a principl

Explore this link on the map →

saved by

related reading