Exploration Of How Large Language Models Navigate Decision Making With Strategic Prompt Engineering And Summarization

Researchers have explored the potential of using Large Language Models (LLMs) in reinforcement learning (RL) environments, particularly in multi-armed bandit (MAB) problems. Results show that LLMs have limited exploration capabilities without specific interventions, but a specially designed prompt can lead to satisfactory exploratory behavior.