Search papers, labs, and topics across Lattice.
5
0
8
6
SkillAdaptor achieves targeted skill updates that enhance LLM performance by over 1.5 points on key benchmarks, revolutionizing how agents adapt to failures in real-time.
AbaqusAgent turns natural language into successful finite element analyses, achieving an 86% success rate and transforming how engineers interact with simulation tools.
Latent-based collaboration may obscure attack risks, with latent-only attacks capable of severely degrading performance without any visible adversarial text.
Industry-standard CPU benchmarks can vary by up to 436% due to uncontrolled configurations of other system components, highlighting a critical flaw in current evaluation practices.
AI agents can now learn durable skills instead of constantly "reinventing the wheel," thanks to SkillNet's infrastructure for creating, evaluating, and connecting AI skills at scale.
