From hard refusals to safe-completions: toward output-centric safety training

2025-08-07来源 OpenAI BlogEnglish
Discover how OpenAI's new safe-completions approach in GPT-5 improves both safety and helpfulness in AI responses—moving beyond hard refusals to nuanced, output-centric safety training for handling dual-use prompts.

原文：https://www.a8q9.cn/en/article/from-hard-refusals-to-safe-completions-toward-output-centric-safety-training-f1c3dc.html

相关文章