Search papers, labs, and topics across Lattice.
1
0
2
Factorizing world states with language unlocks surprisingly strong zero-shot reward prediction across diverse environments, outperforming end-to-end learned critics and LLM-as-a-judge approaches.
