Search papers, labs, and topics across Lattice.
Columbia University
5
0
8
2
Latent communication can be compressed by 75% without accuracy loss by conditioning KV cache transmissions on receiver information needs.
Transforming data systems from passive repositories into active agents could redefine the landscape of autonomous automation and its safety protocols.
Even state-of-the-art LLMs like GPT-5.2 falter in LakeQA, scoring just 18.37% on a benchmark that demands both searching and multi-hop reasoning.
VISTA reveals that integrating UI and API interactions can drastically enhance the realism and comprehensiveness of agent evaluations, outperforming existing benchmarks.
Reduce your MoE LLM's memory bottleneck by up to 49% with a simple router fine-tuning trick that encourages experts to be reused.
