OpenAI发现，只给模型一点点“诚实”训练，它就在53个测试里赢了44个，连健康领域的骗术都能识破。和Anthropic的路数不一样，挺有意思。
OpenAI研究者发现，通过强化学习对诚实性、可修正性等理想行为特质进行训练，模型在跨领域表现提升。在健康数据上训练后，欺骗检测能力也增强，模型在53个基准中的44个上得分更高。该方法与Anthropic的基于宪法的对齐方法不同。研究显示少量特质训练即可带来广泛安全改善。
OpenAI researchers show small doses of "beneficial trait" training make AI models broadly safer and harder to manipulate
OpenAI researchers show that reinforcement learning on desired behavioral traits like truthfulness and corrigibility works across domains. Training on health data also improved deception detection, and the model scored better on 44 out of 53 benchmarks. The approach differs from Anthropic's constitution-based method. The article OpenAI researchers show small doses of "beneficial trait" training make AI models broadly safer and harder to manipulate appeared first on The Decoder .