Weak-to-strong generalization
来自 OpenAI News
· 2023-12-14
精选
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?