OpenAI News·· 2023-12-14AI 评分47
OpenAI 提出弱到强泛化以控制强模型
Weak-to-strong generalization
AI 导读
OpenAI 提出“弱到强泛化”这一超级对齐新研究方向，旨在利用深度学习的泛化特性实现用弱监督者控制强模型。该研究展示了初步的积极结果，探索了通过弱监督信号引导更强 AI 系统的可能性。
原文
We present a new research direction for superalignment, together with promising initial results: can we leverage the generalization properties of deep learning to control strong models with weak supervisors?
来源：OpenAI News · openai.com