WebFactory: Automated Compression of Foundational Language Intelligence into Grounded Web Agents

WebFactory introduces a fully automated reinforcement learning pipeline that efficiently transforms large language models into GUI agents without requiring unsafe live web interactions or costly human-annotated data. The system demonstrates exceptional data efficiency by achieving comparable performance to human-trained agents while using synthetic data from only 10 websites.

• →WebFactory eliminates the need for unsafe live web interactions or expensive human-crafted training data for GUI agents.
• →The system achieves comparable performance to human-trained agents using synthetic data from only 10 websites.
• →The pipeline includes scalable environment synthesis, knowledge-aware task generation, and decomposed reward RL training.
• →The research provides insights into 'embodiment potential' as a new axis for evaluating LLM foundations.
• →This approach offers a scalable paradigm for converting passive internet knowledge into active, grounded AI intelligence.