Know What You Don't Know: Agentic Data Synthesis by Evolving Generation Space for Closed-Loop Learning

Modern Artificial Intelligence is often measured by speed and accuracy, yet a critical challenge remains: the tendency for systems to provide confident but incorrect responses. Recognizing when not to reply is becoming a key sign of intelligence. To address learning gaps in interactive agents, EvoSyn is proposed as a learning-feedback-driven multi-agent framework. This approach focuses on Agentic Data Synthesis by evolving generation spaces for closed-loop learning, ensuring synthetic experiences provide genuine learning value for task agents.