BeClaude
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