False faced: can a simple classifier tell when a model is manipulating its activations to evade probes?

This experiment was a submission to my Statistics & Probability course Bayesian classifiers project. TL;DR: We test whether simple classifiers can tell when a model is hiding concepts in its activations to evade probes. We find that, for Bayesian classifiers, the probabilities are enough to separate most evading rows from the rest on concepts seen in training, but the correlation aspect through the pairs of dimensions that the Dependent version adds over the Naive approach does not measurably help. None of the classifiers work on a new concept, because they learn the trigger word that elicits the evasion, rather than the actual evasion, which suggets the results on the seen concepts may mostly be based on the trigger words. Basics: probes and activations Everybody lies * . Humans can say one thing while having something completely different in their mind, and language models can do it too. For a model to generate a response, information must flow downstream through a stack of layers (42 for the model we use in this experiment). Models don't process standard text, they process tokens. The text gets split into short pieces of text (tokens), each assigned a vector (a list of numbers,