Framework to Optimize LLM Tool Calling Decisions
A new framework from arXiv (2605.00737) addresses the decision of whether LLMs should call external tools, particularly web search. Inspired by decision-making theory, it evaluates tool-use along necessity, utility, and affordability. The analysis combines normative and descriptive perspectives to infer true need versus self-perceived need. The work aims to optimize agentic AI architectures by reducing redundant or harmful tool calls.
Key facts
- arXiv paper 2605.00737 introduces a framework for LLM tool calling decisions.
- The framework focuses on web search tools.
- Three key factors: necessity, utility, and affordability.
- Combines normative and descriptive perspectives.
- Aims to reduce redundant or harmful tool calls.
- Published on arXiv as a new announcement.
- Title: 'To Call or Not to Call: A Framework to Assess and Optimize LLM Tool Calling'.
- Addresses agentic AI architectures augmenting LLMs with external tools.
Entities
Institutions
- arXiv