“How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questions”, Lorenzo Pacchiardi, Alex J. Chan, Sören Mindermann, Ilan Moscovitz, Alexa Y. Pan, Yarin Gal, Owain Evans, Jan Brauner2023-09-26 (GPT calibration, AI safety)⁠:

Large language models (LLMs) can “lie”, which we define as outputting false statements despite “knowing” the truth in a demonstrable sense. LLMs might “lie”, for example, when instructed to output misinformation.

Here, we develop a simple lie detector that requires neither access to the LLM’s activations (black-box) nor ground-truth knowledge of the fact in question. The detector works by asking a predefined set of unrelated follow-up questions after a suspected lie, and feeding the LLM’s yes/no answers into a logistic regression classifier.

Despite its simplicity, this lie detector is highly accurate and surprisingly general. When trained on examples from a single setting—prompting GPT-3.5 to lie about factual questions—the detector generalizes out-of-distribution to (1) other LLM architectures, (2) LLMs fine-tuned to lie, (3) sycophantic lies, and (4) lies emerging in real-life scenarios such as sales.

These results indicate that LLMs have distinctive lie-related behavioral patterns, consistent across architectures and contexts, which could enable general-purpose lie detection.