claim
active
claim:datapoint-features-are-not-mechanistically-faithful-when-the-underlying-model-has-no-special-notion-of-p

Datapoint features are not mechanistically faithful when the underlying model has no special notion of p

Memorization by transcoder not in model.

Source paper

extracted_from
A Toy Model of Mechanistic (Un)Faithfulness
Chris Olah

Neighborhood — ranked by edge-count