Weak-to-strong generalization
✨ AI Summary
🔊 جاري الاستماع
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?
