This AI Paper Proposes an Interactive Agent Foundation Model that Uses a Novel Multi-Task Agent Training Paradigm for Training AI Agents Across a Wide Range of Domains, Datasets, and Tasks