Alibaba and ZJU Propose TTPO for Label-Free Test-Time Training
burkov · x · 2026-08-29
A paper from Alibaba Group and Zhejiang University proposes Test-Time Policy Optimization (TTPO), an asymmetric framework that distills agreeing rollouts while penalizing disagreeing rollouts under majority-vote pseudo-labels. This approach achieves label-free test-time training that matches ground-truth supervised distillation performance across competition-level mathematical benchmarks.
More from Research
- Interpretability paper incoming: in-depth look at testing explanations on modern LLMs — burny_tech · 2026-10-16
- Better-learning models tend to be more interpretable, argues Nick Cammarata — burny_tech · 2026-10-16
- Cornell and Amazon revisit the 16x pretraining compute gap: diffusion LMs need only 3-8x — LucaAmb · 2026-10-16
- AI trained on the fraction of human intelligence that fit the page — r0ck3t23 · 2026-10-16
- Quotient Recurrent Reasoners hit 97.9% on Sudoku-Extreme with 78% fewer loops — burny_tech · 2026-10-16
- Mathematicians publish explainer for the Egyptian fraction proof in OpenAI's proofs collection — burny_tech · 2026-10-16