Search papers, labs, and topics across Lattice.
5
1
9
33
Diverse LLMs, when strategically routed and collaborating, can collectively achieve significantly fairer outputs than any single model acting alone.
Mutation analysis as an adequacy metric for kernel-benchmark oracles is introduced: deterministic rules inject 10 compilable faults into verified CUDA implementations of 188 KernelBench problems, 7{,}384 of them with an independent kill witness; any test protocol is scored by the fraction it detects.
Named-material rendering is largely solved, but grounding materials from pixels remains an open challenge, as evidenced by significant performance drops with reference-image conditioning.
Self-correction in LLMs can be dramatically improved by reinforcing step-level reasoning, achieving higher accuracy and reliability in outputs.
Learning high-level strategies can boost vulnerability reproduction success rates by over 20%, revolutionizing how we approach software security tasks.
