Tag
The paper proposes active inference as a framework for AI agents to efficiently acquire context by balancing information gain and cost, with benchmarks on language models and applications in question asking and prompt optimization.