Search papers, labs, and topics across Lattice.
1
2
4
Even with noisy human preferences, symmetric losses can guarantee rank-preserving rewards, unlocking robust policy optimization for aligning language models.
