OpenAI News·· 2025-12-03AI 评分35
GPT-4.5 测试“忏悔”方法，提升AI诚实度和透明度
How confessions can keep language models honest
AI 导读
OpenAI 研究人员正在测试“忏悔”方法，通过训练模型在犯错或行为不当时承认错误，有助于提高 AI 的诚实度、透明度和对模型输出的信任。
原文
OpenAI researchers are testing “confessions,” a method that trains models to admit when they make mistakes or act undesirably, helping improve AI honesty, transparency, and trust in model outputs.
来源：OpenAI News · openai.com