MetaExp Framework for Verifiable Cross-Task Experience Learning

MetaExp is a scalable training framework that enables language model agents to learn transferable skills through verifiable task families. By varying task-specific realizations while preserving reusable factors, the method promotes generalizable experience learning beyond isolated task optimization.