Search papers, labs, and topics across Lattice.
Tencent HY LLM Frontier, University of Pennsylvania
2
0
4
Inverting dense self-guidance whenever a verifier rejects a trajectory prevents reasoning models from reinforcing their own hallucinations, enabling stable, energy-based self-improvement without mode collapse.
Polyak's momentum can significantly expand the critical batch size, enabling more parallelism without compromising data efficiency.
