Beyond External Monitors: Enhancing Transparency of Large Language Models for Easier Monitoring

TL;DR

This paper introduces TELLME, a novel approach to improve the transparency of large language models by making their internal processes more interpretable, thereby enhancing monitoring and trustworthiness in safety-critical applications.

Contribution

The paper proposes TELLME, a new method that enhances LLM transparency and monitoring capabilities, with theoretical analysis of its impact on generalization using optimal transport theory.

Findings

TELLME improves transparency in LLMs.

TELLME enhances safety and detoxification task performance.

Theoretical analysis shows improved generalization ability.

Abstract

Large language models (LLMs) are becoming increasingly capable, but the mechanisms of their thinking and decision-making process remain unclear. Chain-of-thoughts (CoTs) have been commonly utilized to monitor LLMs, but this strategy fails to accurately reflect LLMs' thinking process. Techniques based on LLMs' hidden representations provide an inner perspective to monitor their latent thinking. However, previous methods only try to develop external monitors instead of making LLMs themselves easier to monitor. In this paper, we propose a novel method TELLME, improving the transparency of LLMs and helping monitors identify unsuitable and sensitive behaviors. Furthermore, we showcase the applications of TELLME on trustworthiness tasks (\eg, safety risks monitoring tasks and detoxification tasks), where LLMs achieve consistent improvement in transparency and task performance. More crucially, we theoretically analyze the improvement of TELLME on LLMs' generalization ability through optimal transport theory.