Search papers, labs, and topics across Lattice.
University of Maryland
1
0
2
Scoring fixed-length reasoning chunks with LLMs can outperform traditional majority voting by up to 28 percentage points, all without the need for reward model training.
