Interpreting Model Reasoning through Behavioral Counterfactuals
Holds a model response fixed while changing the prompt, then learns hidden-state edits that reproduce the change in token probabilities. Uses these edits and activation patching to show how suggestions can influence reasoning through question and filler-token states without being mentioned in the written response.
First public: · Added to this list: October 23, 2026
Explore this approach
Analysis and evaluation — Measure how well these methods work, investigate what their internal computation does, and identify where they fail. This includes experiments, studies of model internals, and mathematical results.