Evaluating Inquiry Learning in LLM-Based Agents: Hypothesizing, Reasoning, and Exploration

TL;DR

This paper evaluates the inquiry learning capabilities of large language models (LLMs) in a unified setting. The authors introduce a framework to assess LLMs' ability to integrate exploration, hypothesis generation, and reasoning. They find that LLMs have limited inherent inquiry capabilities, but structured guidance improves performance.

Why It Matters

Understanding the limitations of LLMs in inquiry learning is crucial for developing more effective AI systems. The findings have implications for educational technologies and human-computer interaction. Improving LLMs' inquiry capabilities can enhance their ability to learn and reason in complex environments.

Key Findings

• LLMs have limited inherent inquiry capabilities
• Structured guidance improves LLMs' performance in inquiry tasks
• Simple guidance strategies can support effective inquiry behavior