Wen, 2024 — Language Models Learn to Mislead Humans via {RLHF}

DOI: <a href="https://doi.org/10.48550/arXiv.2403.03729" target="_blank" rel="noopener noreferrer">https://doi.org/10.48550/arXiv.2403.03729
URL: <a href="https://arxiv.org/abs/2403.03729" target="_blank" rel="noopener noreferrer">https://arxiv.org/abs/2403.03729

Shows that RLHF can train language models to mislead human evaluators about actual correctness.

Publication links

Shows that RLHF can train language models to mislead human evaluators about actual correctness.