Google's OPST: LLM self-improvement without external data

Google and Ulm University propose On-Policy Self-Training, letting LLMs generate their own questions and reasoning traces to improve without external data, though self-generated training data risks diversity collapse, hurting medical QA performance.

2026-10-26 ~ 2026-10-27 · 3 related posts

• Google Paper: Data-Free On-Policy Self-Training Improves LLMs Across Text and Vision — google · 2026-10-26
• OPST paper shows LLMs self-improve with zero human data, guided only by topic descriptions — _akhaliq · 2026-10-26
• On-Policy Self-Training paper: self-generated training data causes diversity collapse — idanbeck · 2026-10-27