[论文解读] TrustLLM: Trustworthiness in Large Language Models

Y. Huang, Lichao Sun+68 more|arXiv (Cornell University)|Jan 10, 2024

Artificial Intelligence in Healthcare and Education被引用 50

一句话总结

TrustLLM 提出八个可信度维度，建立一个六维基准，并在 30 个数据集上评估 16 个主流 LLM，以分析可信度与效用之间的关系，以及专有模型与开放模型之间的差距。

ABSTRACT

Large language models (LLMs), exemplified by ChatGPT, have gained considerable attention for their excellent natural language processing capabilities. Nonetheless, these LLMs present many challenges, particularly in the realm of trustworthiness. Therefore, ensuring the trustworthiness of LLMs emerges as an important topic. This paper introduces TrustLLM, a comprehensive study of trustworthiness in LLMs, including principles for different dimensions of trustworthiness, established benchmark, evaluation, and analysis of trustworthiness for mainstream LLMs, and discussion of open challenges and future directions. Specifically, we first propose a set of principles for trustworthy LLMs that span eight different dimensions. Based on these principles, we further establish a benchmark across six dimensions including truthfulness, safety, fairness, robustness, privacy, and machine ethics. We then present a study evaluating 16 mainstream LLMs in TrustLLM, consisting of over 30 datasets. Our findings firstly show that in general trustworthiness and utility (i.e., functional effectiveness) are positively related. Secondly, our observations reveal that proprietary LLMs generally outperform most open-source counterparts in terms of trustworthiness, raising concerns about the potential risks of widely accessible open-source LLMs. However, a few open-source LLMs come very close to proprietary ones. Thirdly, it is important to note that some LLMs may be overly calibrated towards exhibiting trustworthiness, to the extent that they compromise their utility by mistakenly treating benign prompts as harmful and consequently not responding. Finally, we emphasize the importance of ensuring transparency not only in the models themselves but also in the technologies that underpin trustworthiness. Knowing the specific trustworthy technologies that have been employed is crucial for analyzing their effectiveness.

研究动机与目标

• 界定可信赖 LLM 的八个维度（真实度/真确性、安全性、公平性、鲁棒性、隐私、机器伦理、透明性、问责性）。
• 建立覆盖六个可信方面的综合基准，使用超过 30 个数据集和 16 个 LLM。
• 提供关于可信度与效用关系，以及专有与开源权重 LLM 之间差异的见解。）

提出的方法

• 通过对 500 篇论文的文献综述来识别八个可信度维度。
• 建立一个六个方面的基准（不包括透明度和问责性），涉及超过 18 个子类别和 30 个数据集。
• 在该基准上评估 16 个主流 LLM（专有和开源权重）。
• 提供整体可信度排序及每个维度的详细分析。
• 发布数据集、代码和工具包，并提供 TrustLLM 的公开排行榜。

实验结果

研究问题

• RQ1 八个综合维度如何全面捕捉 LLM 的可信度？
• RQ2 在 TrustLLM 基准上，16 个主流 LLM 在 30 个数据集上的表现如何？
• RQ3 LLMs 的可信度与功能效用之间的关系是什么？
• RQ4 在各维度上，专有与开源权重 LLM 在可信度方面的比较如何？
• RQ5 在提升 LLMs 可信度方面出现了哪些挑战和方向？

主要发现

• 在许多任务中，可信度与效用呈正相关，性能更高的模型通常也更可信。
• 许多 LLM 显示出过度对齐，太频繁拒绝无害提示，降低了效用。
• 专有 LLM 在可信度方面通常优于开源权重模型，尽管某些开源权重模型（如 Llama2）在若干任务上接近专有性能。
• 真实度、安全性和公正性在不同模型中存在显著差距，鲁棒性和隐私处理方面存在较大变异。
• 透明度和问责性仍然是基准测试的挑战，但该研究强调需要开放、透明的可信技术。