Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
1
Fusing hidden-state representations from multiple reward models can dramatically improve the evaluation of non-verifiable tasks, achieving superior performance over traditional methods.