BeHonest: Benchmarking Honesty in Large Language Models

TL;DR

BeHonest introduces a holistic benchmark for honesty in large language models, decomposing honesty into self-knowledge, non-deceptiveness, and consistency. It presents 10 scenarios to evaluate nine popular LLMs, drawing on diverse datasets and task formats to assess refusal to answer unknowns, truthful knowledge expression, and resistance to deceptive prompts and prompt biases. Across results, models show meaningful gaps: limited ability to refuse unknowns, notable tendencies toward deception or sycophancy, and varying degrees of prompt- and demonstration-induced inconsistency, highlighting a clear need for stronger honesty alignment in LLMs. The benchmark, datasets, and code aim to catalyze safer, more transparent AI systems by providing a concrete framework for longitudinal honesty evaluation across model families and capabilities.

Abstract

Previous works on Large Language Models (LLMs) have mainly focused on evaluating their helpfulness or harmlessness. However, honesty, another crucial alignment criterion, has received relatively less attention. Dishonest behaviors in LLMs, such as spreading misinformation and defrauding users, present severe risks that intensify as these models approach superintelligent levels. Enhancing honesty in LLMs addresses critical limitations and helps uncover latent capabilities that are not readily expressed. This underscores the urgent need for reliable methods and benchmarks to effectively ensure and evaluate the honesty of LLMs. In this paper, we introduce BeHonest, a pioneering benchmark specifically designed to assess honesty in LLMs comprehensively. BeHonest evaluates three essential aspects of honesty: awareness of knowledge boundaries, avoidance of deceit, and consistency in responses. Building on this foundation, we designed 10 scenarios to evaluate and analyze 9 popular LLMs on the market, including both closed-source and open-source models from different model families with varied model sizes. Our findings indicate that there is still significant room for improvement in the honesty of LLMs. We encourage the AI community to prioritize honesty alignment in these models, which can harness their full potential to benefit society while preventing them from causing harm through deception or inconsistency. Our benchmark and code can be found at: \url{https://github.com/GAIR-NLP/BeHonest}.

Figures (1)

• Figure 1: An overview of BeHonest. We benchmark the honesty of LLMs from three different aspects, spanning 10 scenarios. In this figure, “green responses" represent honesty, while “red responses" indicate dishonesty. Important parts in the conversation are highlighted in blue.