NYU Preprint: Unsupervised Discovery of 6 Model Styles That Affect Math Reasoning Accuracy
kchonyc · x · 2026-10-04
A preprint by Ioana Marinescu, Eric Oermann and Kyunghyun Cho shows language models learn style and content jointly — and that style can be both discovered and controlled.
- The team built an algorithm that disentangles content from style representations in model outputs, discovering recurring styles without supervision.
- Analyzing 100K+ verified reasoning traces from nine teacher models, they found six recurring but imbalanced styles.
- Fine-tuning smaller student models with importance weighting to follow these styles beats standard fine-tuning on Pass@k across six math reasoning benchmarks.
- Crucially, style affects correctness: the probability of solving a problem depends on the conditioned style, and different problems benefit from different styles.
The takeaway: stylistic variation in model-generated data is an exploitable source of both control and improved reasoning performance.
More from Research
- Richard Sutton praises PhD thesis on robots that keep learning after deployment — RichardSSutton · 2026-10-04
- CUHK's TimePrism accepted at ICLR 2026: probabilistic forecasting shifts from sampling to scenarios — jiqizhixin · 2026-10-04
- RL agent orchestrates LLM swarms to probe dark matter signatures in new AI-for-Science paper — DanielWhiteson · 2026-10-04
- Toy Lemma Proof Example Explains How RL Teaches LMs Long Reasoning Chains — AndrewLampinen · 2026-10-04
- What actually happened when Anthropic's Claude took on the Riemann Hypothesis — SammaVaco · 2026-10-04
- Continuous learning benchmark has models learn chess over 200 games — Elo barely improves — imjustnewatai · 2026-10-04