OpenAI releases LASER: recursive sampling to filter safety evaluation dialogues
Meet LASER: Recursive sampling for safer conversations
AI summary
OpenAI introduces the LASER method, selecting samples near the safety policy boundary from synthetic and de-identified conversations to build evaluation sets covering rare scenarios and reduce over-refusals. The authors state that this method requires significantly less annotation compute than random sampling. This efficiency result does not itself imply that models have become safer.
Why it was selected
OpenAI 公开了 LASER 采样管线的完整设计，包括逻辑回归采样、多样性选择与隐私约束，可供构建安全评测集的团队参考。