Search papers, labs, and topics across Lattice.
3
0
6
0
LLMs that dominate in strategic reasoning often choke in real-time zero-sum games, revealing a critical strategy-execution gap that current benchmarks miss.
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.
Low-light image enhancement gets a boost: BiEvLight's bi-level learning approach tackles noise coupling in event data, leading to substantial improvements over existing methods.
