Reinforced Internal-External Knowledge Synergistic Reasoning for Efficient Adaptive Search Agent
Ziyang Huang, Xiaowei Yuan, Yiming Ju, Jun Zhao, Kang Liu
2025-05-13
Summary
This paper talks about the IKEA agent, an AI system that gets better at searching for information and answering questions by smartly combining what it already knows with new facts it finds online or in databases.
What's the problem?
The problem is that language models sometimes make things up, called hallucinations, or waste time looking for information they already know. This makes them less reliable and less efficient when helping people find answers.
What's the solution?
The researchers designed the IKEA agent to use both its own built-in knowledge and outside sources in a balanced way. By reinforcing this approach with special training, the agent learns when to trust what it knows and when to look things up, which cuts down on mistakes and saves time.
Why it matters?
This matters because it makes AI assistants more trustworthy and efficient, helping people get accurate information faster and with fewer errors, which is useful for research, studying, and everyday problem-solving.
Abstract
The IKEA agent optimizes reinforcement learning in LLMs by synergistically using internal and external knowledge, reducing hallucinations and unnecessary retrievals.