「Multi-SWE-RL is an open-source community aimed at developing high-quality RL training datasets for complex software engineering tasks. Its purpose is to serve as the foundational infrastructure for training fully autonomous agents capable of addressing real-world software engineering challenges, paving the way toward achieving AGI.」とAGIに言及があるのと「In light of these advancements, we are firmly convinced that “scaling RL in real-world environments is the path toward human-like intelligence”.」は熱い。