Activation probing
A method for detecting model states or attributes by analyzing patterns in the model's internal activations.
Timeline (last 90 days)
On the four datasets of FaithCoT-Bench and BonaFide, VERITAS's balanced accuracy improves on average by 16.0% over activation probing.
Activation probing trains a linear classifier on the activations of one or several layers of the model to predict faithfulness; it is a static, single-point detection that struggles to capture inter-layer dynamics.
All Facts (2)
On the four datasets of FaithCoT-Bench and BonaFide, VERITAS's balanced accuracy improves on average by 16.0% over activation probing.
50%UnverifiedActivation probing trains a linear classifier on the activations of one or several layers of the model to predict faithfulness; it is a static, single-point detection that struggles to capture inter-layer dynamics.
50%