Search papers, labs, and topics across Lattice.
4
0
9
SkillComposer achieves a remarkable +23.1% increase in task success rates for LLM agents by rethinking how skills are composed and executed together.
AI peer reviewers are alarmingly vulnerable to targeted multimodal attacks, with existing defenses falling short in the face of sophisticated adversarial strategies.
Achieve SOTA monocular 3D reconstruction and novel view synthesis without sacrificing speed, by fusing Gaussian Splatting with voxel-based SDFs.
Spotting unfaithful reasoning in LLMs just got easier: a new method efficiently compares a model's internal computations against its stated rationale.
