Search papers, labs, and topics across Lattice.
The most-read papers, busiest labs, and emerging topics - updated every morning.
We read everything so you don't have to. One Lattice AI email, no fluff.
Token-level LLM routing, previously hindered by system inefficiencies, now achieves up to 64x higher throughput with a novel request-centric, model-centric serving architecture.
→ This work unlocks substantial cost-efficiency and quality improvements for LLM serving by enabling fine-grained routing strategies at scale.
A single filter can reduce unsafe motion generation by 90% by grounding safety rules in the generator's own reference trajectories, without retraining.
→ This work provides a practical, training-free mechanism to imbue generative motion models with scene-aware safety, crucial for their deployment in human-robot interaction.
One million parallel environments are insufficient without intelligent sampling: Success Guided Sampling unlocks mega-scale RL for robots by focusing training on the edge of learned capabilities.
→ This work provides a scalable and effective approach to sim-to-real RL for complex robotic tasks, reducing reliance on manual engineering and enabling broader generalization.
LLMs can be steered to reliably interpret complex genomic variants by grounding their reasoning in deterministic biological computations and evidence verification, avoiding hallucinated TF binding changes and fabricated experimental support.
→ This work offers a robust framework for unlocking the functional significance of vast numbers of noncoding variants, accelerating genomic medicine discovery.
Current multimodal agents can build sophisticated perception-control pipelines but fail to reliably execute physical tasks, losing object states and mistaking unfinished goals for completion.
→ This benchmark provides a critical empirical foundation for identifying and addressing the significant capability gaps preventing current AI agents from reliably operating in the physical world.
Rolling 7-day digest, refreshed daily after ingestion at 06:00 UTC
