Search papers, labs, and topics across Lattice.
2
0
6
1
Factually dubious LLM outputs can be tamed by strategically penalizing high-confidence predictions at "risky" tokens during fine-tuning, guided by sentence-level factuality labels.
LLM-as-a-judge consensus is often an illusion: models agree on surface-level features, but diverge wildly when evaluating true quality, a problem fixable by injecting domain knowledge into rubrics.
