Search papers, labs, and topics across Lattice.
6
0
7
5
Query-conditioned reuse boosts agent success by 10.7 points while slashing token usage by nearly 50%, transforming how we leverage past experiences in AI tasks.
Even the best vision-language models struggle with reliable evaluation of computer-using agents, but OS-Shepherd models offer a low-cost solution that matches their performance.
Converting noisy, human-centric guides into self-evolving agent skills can yield performance improvements of up to 25.3 percentage points across diverse tasks.
Even the best multimodal models struggle to understand dynamic charts, achieving only 84.5% accuracy on the new ChartAct benchmark that requires interaction to reveal key information.
Coding agents can now evolve their own harnesses to outperform human-designed ones, thanks to a novel observability-driven approach.
OpenMobile proves that high-performing mobile agents can be trained on entirely synthetic, open-source data, closing the gap with closed-source models and enabling broader research.
