Search papers, labs, and topics across Lattice.
7
0
8
5
RedAct can cut procedural capability leakage from agent traces by over 44% while preserving critical audit evidence.
OPD's unique update geometry reveals that it operates in a low-dimensional channel, fundamentally altering our understanding of model training dynamics.
LLMs struggle with adaptive planning, achieving only 67.75% accuracy when faced with progressively revealed world and user constraints.
Forget RLHF, denoising feedback offers a surprisingly effective and scalable alternative for training diffusion language models.
Current multimodal agents still struggle to combine ambiguous visual cues with open-web verification, highlighting a critical gap in their ability to perform complex geolocation tasks.
Multimodal agents can now continually improve their tool use and orchestration in open-ended settings without parameter updates, thanks to a novel dual-stream framework that learns from both past experiences and structured skills.
Even the best multimodal agents struggle with realistic visual scenarios, achieving only 27% accuracy on the new AgentVista benchmark that demands long-horizon tool use across web search, image search, and code.
