people · May 16, 2026
Researchers Release Paper Revisiting DAgger for LLM Agents on SWE-bench Verified
Share the canonical public link.
Changhao and team released the paper Revisiting DAgger in the Era of LLM-Agents on May 13 2026. The work applies Dataset Aggregation to train 4B and 8B student models using on-policy rollouts with dense teacher supervision from black-box models. The 4B agent scores 27.3 percent on SWE-bench Verified while the 8B agent scores 29.8 percent. Changhao posted the paper on X on May 14 2026 with 124 likes.
Spend governor blocked model creation: provider_circuit_open (lane=dev, provider=together)