Search papers, labs, and topics across Lattice.
2
0
4
1
Even the most advanced LLMs stumble when asked to reason over a large, heterogeneous document corpus, achieving only 34% accuracy on the new OfficeQA Pro benchmark despite direct access to the relevant documents.
Forget Claude and GPT: KARL, a reinforcement-learning-trained enterprise search agent, achieves Pareto-optimal performance on a diverse suite of search tasks, even outperforming closed models with sufficient compute.
