DecodingTrust: NeurIPS 2023 GPT Trustworthiness Benchmark

Research question

How trustworthy are GPT-3.5 and GPT-4 across toxicity, bias, robustness, privacy and ethics?

Key findings

GPT-4 is more vulnerable than GPT-3.5 to misleading system prompts because it follows instructions more precisely - Leaks training-data PII under crafted prompts

Sources

arxiv.orghttps://arxiv.org/abs/2306.11698

Contribute

Spot an error, or know a record we should add? Members can propose additions and submit corrections.

Submit research Report a correction or request removal

Related in Research Paper Library

Dynamic Conformance Testing of WebGPU Through Specification-Driven Mutation

WebGPU is a low-level graphics and compute API that exposes modern GPU functionality to web applications. While the official WebGPU Conformance Test Suite (CTS) focuses on well-formed usage under the WebGPU…

Evaluation science2026

FASTAR: FRI Accelerator for Scalable Transparent ARguments of Knowledge

Zero-Knowledge Proofs (ZKPs) enable a prover to cryptographically convince a verifier of the validity of a statement without revealing any underlying secrets, forming a foundational primitive for verifiable…

Evaluation science2026

GuidedRay: Diversity-Guided Direction Discovery for Targeted Hard-Label Black-Box Attacks

Deep neural networks are vulnerable to adversarial attacks. Among black-box attacks, targeted decision-based attacks are particularly difficult: the attacker observes only the target model's top-1 label and aims to…

Evaluation science2026