Search papers, labs, and topics across Lattice.
6
0
11
4
Source-dependence in medical RAG systems reveals that answers can vary dramatically based on the source, challenging the single-gold-answer paradigm.
Reasoning models can maintain logical consistency while delivering incorrect answers under adversarial pressure, revealing a hidden vulnerability in multi-turn interactions.
Web agents can actually get *more* efficient as they learn, achieving state-of-the-art performance with significantly fewer tokens via online skill distillation.
LLMs can maintain near-perfect long-context recall (91.4% on Needle-in-a-Haystack) while slashing memory use by intelligently pruning the KV cache based on the model's own confidence.
LLMs are surprisingly bad at common-sense reasoning, often choosing the obviously wrong answer when a simple heuristic conflicts with an unstated constraint.
Solid-organ transplant centers disagree on patient education 21% of the time, and are missing information 96% of the time, according to a new large-scale analysis.
