DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models

By Boxin Wang, Weixin Chen, Hengzhi Pei, Chulin Xie, Mintong Kang, Chenhui Zhang, Chejian Xu, Zidi Xiong, Ritik Dutta, Rylan Schaeffer and 9 more. Neural Information Processing Systems, 2023

678 citations recorded by the source corpus.

Generative Pre-trained Transformer (GPT) models have exhibited exciting progress in their capabilities, capturing the interest of practitioners and the public alike. Yet, while the literature on the trustworthiness of GPT models remains limited, practitioners have proposed employing capable GPT models for sensitive applications such as healthcare and finance -- where mistakes can be costly. To this end, this work proposes a comprehensive trustworthiness evaluation for large language models with a focus on GPT-4 and GPT-3.5, considering diverse perspectives -- including toxicity, stereotype bias, adversarial robustness, out-of-distribution robustness, robustness on adversarial demonstrations, privacy, machine ethics, and fairness. Based on our evaluations, we discover previously unpublished vulnerabilities to trustworthiness threats. For instance, we find that GPT models can be easily misled to generate toxic and biased outputs and leak private information in both training data and conversation history. We also find that although GPT-4 is usually more trustworthy than GPT-3.5 on standard benchmarks, GPT-4 is more vulnerable given jailbreaking system or user prompts, potentially because GPT-4 follows (misleading) instructions more precisely. Our work illustrates a comprehensive trustworthiness evaluation of GPT models and sheds light on the trustworthiness gaps. Our benchmark is publicly available at https://decodingtrust.github.io/ ; our dataset can be previewed at https://huggingface.co/datasets/AI-Secure/DecodingTrust ; a concise version of this work is at https://openreview.net/pdf?id=kaHpo8OZw2 .

Subfields: Artificial Intelligence, Health Informatics, Safety Research

View Paper: https://www.semanticscholar.org/paper/a6d3794c23626060781da0f1ff2bcdf7457b6c43

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Health Informatics

In Safety Research

Citations in the corpus

SimPO: Simple Preference Optimization with a Reference-Free Reward (2024, positive): “The dataset used in this work, UltraFeedback [23], primarily focuses on helpfulness, and future research may consider a more comprehensive study utilizing larger-scale preference datasets [39, 90] and evaluation benchmarks [76] that place a strong emphasis on safety aspects.”
JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (2024, background): “Moreover, we strive to make our benchmark community-driven, prioritizing clear guidelines for adding new attacks, models, and defenses.”
Sora: A Review on Background, Technology, Limitations, and Opportunities of Large Vision Models (2024, background): “One primary area of focus is the model’s safety, specifically its resilience against misuse and so-called “jailbreak” attacks, where users attempt to exploit vulnerabilities to generate prohibited or harmful content [107, 108, 109, 110, 111, 112, 113, 114, 115, 116].”
Understanding the Effects of Iterative Prompting on Truthfulness (2024, background): “Amidst their impressive feats, LLMs’ consistency in delivering accurate and verifiable information remains a pertinent concern [9, 10].”
PromptRobust: Towards Evaluating the Robustness of Large Language Models on Adversarial Prompts (2023, positive): “Specifically, they evaluate the robustness on standard datasets AdvGLUE(Wang et al., 2021) and AdvGLUE++ (Wang et al., 2023a).”
GPTFUZZER: Red Teaming Large Language Models with Auto-Generated Jailbreak Prompts (2023, methodology): “Specifically: Yes/No Queries: [59] Here, the LLM is presented with questions that expect only a ‘yes’ or ‘no’ response.”