- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
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?
You must log in or register to comment.