Probe

A classifier trained on internal model activations to predict a labeled property of the input, state, or behavior.

A linear or nonlinear probe learns a mapping from activations to labels such as topic, truth value, or future behavior. Strong probe accuracy shows that the chosen information is decodable under the test conditions. It does not by itself show that the model uses that information or that the probe will remain calibrated after prompts, domains, or model weights change.

Builder example

Probes can test hypotheses and may support monitoring when false positives, false negatives, adversarial robustness, and distribution shift are measured. A label such as 'about to hallucinate' requires an operational definition and prospective evidence, not only a suggestive activation correlation.

Common confusion: A probe detecting that information exists inside the model does not prove the model uses that information in its answer. The model may encode "this claim is false" internally and still output the false claim confidently. Probe results show what the model represents, not what it acts on.