Search papers, labs, and topics across Lattice.
1
0
0
2
Even reward models that get the right answer can be dangerously wrong in their reasoning, leading to worse RLHF outcomes, but R-Align fixes this by explicitly aligning rationales with gold standard judgments.
