Interpretability and Evals

Attribution graphs, linear probes, capability evaluations, propensity evaluations, and alignment auditing as tools for making model behavior legible.

Attribution graphs, probes, capability evals, and alignment auditing.