Search papers, labs, and topics across Lattice.
4
0
6
OmniAgent's ability to improve performance with more reasoning turns challenges the traditional "watch-it-all" approach in video understanding.
Current audio editing models are failing spectacularly, with an Exact Match Rate below 5% in complex tasks, exposing a critical need for improvement.
Current AI memory systems are surprisingly bad at integrating diverse, real-world information across long time spans, as evidenced by a new benchmark where they only achieve 55% accuracy.
Agent systems leveraging iterative tool orchestration and cross-modal analysis significantly outperform single models in audio reasoning, highlighting a promising path toward explainable audio intelligence.
