Search papers, labs, and topics across Lattice.
4
0
10
0
Multilingual RAG systems are systematically suppressing "answer-critical" documents in non-English languages, crippling their ability to leverage global knowledge.
Forget end-to-end training: breaking down long-context reasoning into atomic skills and training on targeted pseudo-data unlocks a 7.7% performance boost.
Realistic user simulation is now possible: Pare offers a framework that moves beyond flat tool-calling APIs to model stateful user interactions, enabling better evaluation of proactive agents.
MLLMs struggle with visually rendered text not because they can't reason, but because they can't *read* it, and a simple self-distillation fix closes the gap.
