Large Language Models (LLMs) have shown strong performance on tool-use agentic tasks when given a fixed tool schema. Yet a tool schema is not the action spac…
机构:剑桥
来源:arXiv 2609.34971 | AI4Papers 论文推荐平台