Weak-to-strong generalization
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?
14 December 2023 · Source: OpenAI news
The full text is available at the source.