Search papers, labs, and topics across Lattice.
1
0
3
LLMs can achieve state-of-the-art reasoning accuracy with significantly fewer tokens by rewarding intermediate reasoning steps that maximize information gain and maintain monotonic progress.
