Skip to content
LLM Primer

Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned

Authors: Deep Ganguli, Liane Lovitt, Jackson Kernion, Amanda Askell, Yuntao Bai, Saurav Kadavath, Ben Mann, Ethan Perez, Nicholas Schiefer, Kamal Ndousse, Andy Jones, Sam Bowman, Anna Chen, Tom Conerly, Nova DasSarma, Dawn Drain, Nelson Elhage, Sheer El Showk, Stanislav Fort, Zac Hatfield-Dodds, Tom Henighan, Danny Hernandez, Tristan Hume, Joshua Landau, Katherine Lee, Daniel Li, Tom Liao, Chris Olah, Catherine Olsson, Dario Amodei, Tom Brown, Jack Clark, Sam McCandlish, Chris Wallace, Jared Kaplan (2022)

arXiv: 2209.07858

Domains

SafetyEvaluation

TLDR (English)

Systematically studies red teaming methods for language models, finding that harmful output rates may decrease with scale, but models become better at circumventing human-written safety rules. Proposes best practices for scaled red teaming.

TLDR（中文）

系统研究了语言模型的红队测试方法，发现随着模型规模增大，有害输出率反而可能下降，但模型也变得更擅长绕过人类编写的安全规则。提出了规模化红队测试的最佳实践。

Related Papers

Other papers in the same domain