OPST paper shows LLMs self-improve with zero human data, guided only by topic descriptions
_akhaliq · x · 2026-10-26
The On-Policy Self-Training (OPST) paper reframes on-policy self-distillation as a self-training framework: guided only by coarse topic descriptions, the model generates its own question-answer curriculum and privileged reasoning traces, with no human-curated prompts or answers. A pseudo-verification filter keeps problems only if the model reproduces answers in fresh and paraphrased contexts. Experiments cover in-domain self-improvement, out-of-domain knowledge retention, and Visual OPST, showing consistent gains on language and multimodal reasoning benchmarks — evidence that foundation models harbor latent capabilities that can be amplified without external supervision.
Related event: Google's OPST: LLM self-improvement without external data(3 posts)→
More from Research
- EMNLP paper GENIE: a fine-grained metric locating where LLM stories are novel — gregd_nlp · 2026-10-27
- Hashed n-gram value embeddings push GPT to 3.2115 loss, beating 30M track record — RobertTLange · 2026-10-27
- MissionReport benchmark: frontier models score just ~60% on government status reporting — Exp_Mark · 2026-10-27
- TMLR paper: enforced thinking doesn't reliably make LLM agents disclose key info to users — SharonYixuanLi · 2026-10-27
- LLM-written Rust library bbtree verified: speed claims hold, memory savings reverse — lmoroney · 2026-10-27
- Diamandis: AI Cancer Navigator and $100 Genome Sequencing Close the Care Gap — PeterDiamandis · 2026-10-27