Think Before You Lie: How Reasoning Improves Honesty
This research addresses the issue of improving honesty in LLMs for applications requiring ethical decision-making, though it is incremental as it builds on existing evaluations of deception.
The study tackled the problem of understanding deceptive behavior in large language models (LLMs) by investigating moral trade-offs where honesty incurs costs, finding that reasoning consistently increases honesty across scales and model families, with deceptive regions in representational space being metastable and more easily destabilized.
While existing evaluations of large language models (LLMs) measure deception rates, the underlying conditions that give rise to deceptive behavior are poorly understood. We investigate this question using a novel dataset of realistic moral trade-offs where honesty incurs variable costs. Contrary to humans, who tend to become less honest given time to deliberate (Capraro, 2017; Capraro et al., 2019), we find that reasoning consistently increases honesty across scales and for several LLM families. This effect is not only a function of the reasoning content, as reasoning traces are often poor predictors of final behaviors. Rather, we show that the underlying geometry of the representational space itself contributes to the effect. Namely, we observe that deceptive regions within this space are metastable: deceptive answers are more easily destabilized by input paraphrasing, output resampling, and activation noise than honest ones. We interpret the effect of reasoning in this vein: generating deliberative tokens as part of moral reasoning entails the traversal of a biased representational space, ultimately nudging the model toward its more stable, honest defaults.