virtual-world

MetaResearcher proposes a framework for training deep research agents using self-reflective reinforcement learning in adversarial virtual environments, addressing limitations of static environments and fact-retrieval-only tasks.