Back to News
Release2023-12-14
Weak-to-strong generalization
Source: OpenAI
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?
openaigpt