Search papers, labs, and topics across Lattice.
1
0
2
5
Stop reward hacking: disentangling causal and non-causal factors in reward models makes RLHF more robust.
