SPSD distills self-play search records into superhuman CoT, lifting Qwen3-4B math scores
Lorenzo Molfetta · hf · 2026-09-30
SPSD (Self-Play Search Distillation) generates superhuman synthetic reasoning data via self-play of MuZero-like networks on board games, using executable environments to turn search into structured reasoning problems.
- Data generation: The expert identifies a preferred decision, plausible alternatives, plausible opponent replies, and value estimates at each state; search records become environment-grounded chains-of-thought, sidestepping costly human labeling and low-quality synthetic data.
- Transfer: Though trained only on board-game records, SPSD transfers to unseen mathematics.
- Results: On Qwen3-4B-Base, mean score over six math benchmarks rises from 24.1 to 36.6, and held-out-game win rate climbs from 15% to 45%.
Related event: SPSD Paper: Self-Play Search Distillation Boosts LLM Reasoning(3 posts)→
More from Research
- Google Research open-sources PaperOrchestra, a multi-agent framework for writing AI papers — google-research · 2026-11-16
- Kind ICLR Reviewer Raises Score Despite Unresolved Concerns, Citing the Paper's Value — MarkSchmidtUBC · 2026-11-16
- Hand-editing 0.2% of weights plus 100 CPU-minutes turns an English TTS model into an 18MB Sanskrit one — kastnerkyle · 2026-11-16
- Why chirality matters for AI for Science: enantiomers behave differently in biology — madeofAjala · 2026-11-16
- RSS 2027 submissions open with FAQ clarifying two-stage review and negative-result bar — GeorgiaChal · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16