Toward Training Superintelligent Software Agents through Self-Play SWE-RL

N/A

Toward Training Superintelligent Software Agents through Self-Play SWE-RL Yuxiang Wei 12 Zhiqing Sun 3 Emily McMilin 1 Jonas Gehring 1 David Zhang 1 Gabriel Synnaeve 1 Daniel Fried 14 Lingming Zhang 2 Sida Wang 1 Abstract 2025b; FAIR CodeGen team et al., 2025; ByteDance Seed While current software agents powered by large…

saved by

• Ashvin Nair

related reading

• [2512.18552] Toward Training Superintelligent Software Agents through Self-Play SWE-RLarxiv.org
• 2502.18449arxiv.org
• Automated Weak-to-Strong Researcheralignment.anthropic.com
• [2607.00053] SWE-Router: Routing in Multi-turn Agentic Software Engineering Tasksarxiv.org
• [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
• Self-Adapting Language Modelsarxiv.org
• Scaling Self-Play with Self-Guidancearxiv.org
• SWE-1.7: Frontier Intelligence at a Fraction of the Costcognition.com
• SWE-chat: Coding Agent Interactions From Real Users in the Wildarxiv.org
• Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
• [2410.06992] SWE-Bench+: Enhanced Coding Benchmark for LLMsar5iv.labs.arxiv.org
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org