Exploring the generalization of LLM truth directions on conversational formats
This work addresses the challenge of developing reliable LLM lie detectors that can generalize to new conversational settings, though it is incremental as it builds on existing truth direction concepts.
The paper tackled the problem of how truth directions in LLMs generalize across conversational formats, finding good generalization for short conversations ending with a lie but poor generalization for longer formats with earlier lies, and proposed a solution using a fixed key phrase that significantly improved generalization.
Several recent works argue that LLMs have a universal truth direction where true and false statements are linearly separable in the activation space of the model. It has been demonstrated that linear probes trained on a single hidden state of the model already generalize across a range of topics and might even be used for lie detection in LLM conversations. In this work we explore how this truth direction generalizes between various conversational formats. We find good generalization between short conversations that end on a lie, but poor generalization to longer formats where the lie appears earlier in the input prompt. We propose a solution that significantly improves this type of generalization by adding a fixed key phrase at the end of each conversation. Our results highlight the challenges towards reliable LLM lie detectors that generalize to new settings.