Definition
A linear probe is a linear classifier trained on the activations of layer to detect a hidden property of the input: truthfulness vs. lying, factual correctness vs. hallucination, sentiment, topic, bias.
Probes work because the residual stream is additive, so representations are approximately linear. Example: probes on Chess-GPT, trained only on move text, recover the board state and the player’s Elo.
Appears in
- Lecture 2, reading the residual stream
- Lecture 3, System-Level Defenses: probes that detect intent to comply with harm
- Lecture 4, Representation Engineering
- Lecture 7, Applications of Steering and RepE: auditing
- Lecture 10, Deception Probes
- Lecture 10, Test-Awareness Probes
- Lecture 11, Collusion Probes: per-agent deception scores aggregated to a system-level collusion signal