Search papers, labs, and topics across Lattice.
Affiliation:
3
0
3
AGENTSCOPE reveals that integrating structured behavior abstractions with LLM reasoning can dramatically enhance the reliability of diagnosing agent failures.
Skills that are meant to enhance LLM agents can paradoxically lead to significant task failures and inefficiencies, challenging the assumption that more skills always improve performance.
ToolAtlas achieves up to 21.61% better performance in tool utilization by shifting memory management from agents to tool providers, revolutionizing how LLMs interact with external tools.
