OTC: Optimal Tool Calls via Reinforcement Learning
Hongru Wang, Cheng Qian, Wanjun Zhong, Xiusi Chen, Jiahao Qiu, Shijue Huang, Bowen Jin, Mengdi Wang, Kam-Fai Wong, Heng Ji
2025-04-22
Summary
This paper talks about OTC, a new system that teaches language models how to use tools in the smartest and most efficient way possible by using reinforcement learning.
What's the problem?
The problem is that language models often use tools in ways that are not very efficient, which can waste time and resources, and sometimes even cost more money, all while not necessarily giving better answers.
What's the solution?
The researchers created a reinforcement learning framework that helps the model figure out the best times and ways to use different tools. This approach makes the model more efficient and cost-effective, while still keeping its answers accurate.
Why it matters?
This matters because it means AI can do more with less effort and cost, making it more practical for real-world uses like customer service, research, or any job where using digital tools quickly and wisely is important.
Abstract
A novel RL framework optimizes tool usage in language models for enhanced efficiency and cost-effectiveness without sacrificing accuracy.