1
Toward Training Superintelligent Software Agents through Self-Play SWE-RL
通过自我对弈强化学习，让AI在软件工程任务上实现超智能，开创性研究被ICML收录。
arXiv:2512.18552v2 Announce Type: replace-cross Abstract: While current software agents powered by large language models (LLMs) and agentic reinforcem…