Pith. sign in

REVIEW 48 cited by

Beyond Human Data: Scaling Self-Training for Problem-Solving with Language Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2312.06585 v4 pith:CGIBGO63 submitted 2023-12-11 cs.LG

classification cs.LG
keywords datahumanmodelsfeedbackmodelself-trainingbeyondfine-tuning
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
abstract

Fine-tuning language models~(LMs) on human-generated data remains a prevalent practice. However, the performance of such models is often limited by the quantity and diversity of high-quality human data. In this paper, we explore whether we can go beyond human data on tasks where we have access to scalar feedback, for example, on math problems where one can verify correctness. To do so, we investigate a simple self-training method based on expectation-maximization, which we call ReST$^{EM}$, where we (1) generate samples from the model and filter them using binary feedback, (2) fine-tune the model on these samples, and (3) repeat this process a few times. Testing on advanced MATH reasoning and APPS coding benchmarks using PaLM-2 models, we find that ReST$^{EM}$ scales favorably with model size and significantly surpasses fine-tuning only on human data. Overall, our findings suggest self-training with feedback can substantially reduce dependence on human-generated data.

Discussion (0). Sign in to comment.

Forward citations

Cited by 48 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Bridging Compute- and Data-Optimal Pretraining

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Pretraining loss obeys a single law in which repeated or paraphrased tokens count as η(N, data-per-parameter, expansion-ratio) fresh tokens, with total effective data saturating as derived tokens grow.

  2. Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients

    cs.CL 2026-06 unverdicted novelty 7.0 of 10

    ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite w...

  3. Step-by-Step Optimization-like Reasoning in LLMs over Expanding Search Spaces

    cs.AI 2026-06 unverdicted novelty 7.0 of 10

    Introduces OPT* tasks and two training regimes (solver-guided online policy optimization with rank-based reward shaping and search-based offline RL) plus a theoretical link between search success and information extra...

  4. Self-Policy Distillation via Capability-Selective Subspace Projection

    cs.CL 2026-05 unverdicted novelty 7.0 of 10

    Self-Policy Distillation extracts a capability subspace from model gradients on correctness tokens, projects KV activations into it for self-generation, and fine-tunes LLMs to achieve up to 13-16% gains over baselines...

  5. Beyond Negative Rollouts: Positive-Only Policy Optimization with Implicit Negative Gradients

    cs.CL 2026-05 unverdicted novelty 7.0 of 10

    POPO uses bounded importance sampling on positive rollouts and a siamese policy network to achieve implicit negative gradients and stable optimization, matching or exceeding GRPO on math benchmarks such as 36.67% on A...

  6. Turbo Connection: Reasoning as Information Flow from Higher to Lower Layers

    cs.LG 2026-02 conditional novelty 7.0 of 10

    An architecture that routes hidden states from higher to lower layers between consecutive tokens improves reasoning accuracy and length generalization in fine-tuned LLMs.

  7. The Verifier is the Curriculum: Execution-Gated Self-Distillation for Cross-Family Game Generation

    cs.AI 2026-06 conditional novelty 6.5 of 10

    Strict-launch-gated rejection-sampling self-distillation compounds cross-family clean Godot generation from 8.8% to 42.2% and full best-of-K coverage, while gold duplication and a lenient BUILD filter erase the gain.

  8. From Scoring to Acting: Outcome-Verified Comparative Self-Distillation for LLM Agents

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Outcome-verified teacher continuations from student failure prefixes, plus divergence-local comparison and suffix distillation, raise skill-free agent success over scoring-based self-distillation.

  9. Off-Context GRPO: Learning to Reason on Hard Problems using Privileged Information

    cs.LG 2026-07 conditional novelty 6.0 of 10

    OC-GRPO reweights GRPO gradients with an importance ratio so that hints used during rollout generation still optimize the original unguided objective, delivering a 13.8% relative Pass@1 gain over vanilla GRPO.

  10. Answer-Conditioned Chains of Thought Degrade Verifiable-Reasoning Distillation in Large Language Models

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Answer-conditioned chains of thought are harmful supervision for reasoning distillation—causing up to ~27-point accuracy drops—and the damage is invisible to correctness filters.

  11. Post-Training Shifts Confidence: A Three-Stage Analysis of How SFT, RL, and OPD Shape CoT Calibration

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Post-training changes where confidence is trustworthy in reasoning traces: OPD best gauges difficulty before thinking, SFT best supports early stopping, RL best selects final answers, and confidence should be read onl...

  12. UNIBROWSE: A Data-to-Agent Framework for Multimodal BrowseComp

    cs.CL 2026-07 conditional novelty 6.0 of 10

    A unified KG-plus-live-web data pipeline covering all three multimodal BrowseComp information-flow patterns, plus an exploration-degree filter, yields a 35B agent at 54.4 avg accuracy.

  13. Multi-Turn On-Policy Distillation with Prefix Replay

    cs.LG 2026-07 conditional novelty 6.0 of 10

    ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.

  14. Active-GRPO: Adaptive Imitation and Self-Improving Reasoning for Molecular Optimization

    cs.LG 2026-07 unverdicted novelty 6.0 of 10

    Active-GRPO reaches 0.1773 average SRxSim on TOMG-Bench MOLOPT by adaptively switching between imitation and self-reinforcement while upgrading references, outperforming GRPO and RePO.

  15. PHF: Privileged Hidden Flow for On-Policy Self-Distillation

    cs.AI 2026-06 unverdicted novelty 6.0 of 10

    PHF distills token-to-token transition directions and trajectory geometry in hidden states during on-policy self-distillation, reporting 1.5-2.2 point gains on Average@12 for Qwen3-1.7B/4B/8B over reproduced OPSD base...

  16. Paying More Attention to Visual Tokens in Self-Evolving Large Multimodal Models

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    VISE is an unsupervised self-evolving method for LMMs that uses invariance rewards to improve visual conditioning, reporting gains on captioning and reduced hallucination across multiple models.

  17. On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.

  18. DRIFT: Refining Instruction Data via On-Policy Data Attribution

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    DRIFT applies on-policy influence functions with signed weighting and debiasing to attribute and refine SFT data, raising performance on 7B instruction and reasoning models over prior curation methods.

  19. DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    DRIFT achieves multi-turn RL performance via offline importance-weighted SFT by leveraging the equivalence of KL-regularized RL to weighted supervised learning.

  20. FormulaSPIN: Self-Play Fine-Tuning for Natural Language to Spreadsheet Formula Generation

    cs.AI 2026-05 conditional novelty 6.0 of 10

    FormulaSPIN applies execution-aware self-play with an adaptive curriculum to spreadsheet formula generation, improving exact match from 68.2% to 74.9% on NL2Formula-70K without additional data.

  21. RISE: Reliable Improvement in Self-Evolving Vision-Language Models

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    RISE is a self-evolving framework for VLMs that adds fine-grained alternation, quality supervision, and dynamic balancing to produce reliable gains on seven benchmarks from unlabeled data.

  22. Segment-Aligned Policy Optimization for Multi-Modal Reasoning

    cs.AI 2026-05 unverdicted novelty 6.0 of 10

    SAPO introduces segment-level policy optimization using a step-wise MDP abstraction to better align RL updates with reasoning structure in multi-modal LLM tasks.

  23. $S^3$-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data

    cs.LG 2026-05 unverdicted novelty 6.0 of 10

    S^3-R1 generates synthetic intermediate-difficulty multi-hop questions and applies dense rewards for search quality plus answer correctness, yielding up to 10% better out-of-domain generalization than baselines.

  24. Space Syntax-guided Post-training for Residential Floor Plan Generation

    cs.LG 2026-02 unverdicted novelty 6.0 of 10

    SSPT turns space-syntax integration metrics into post-training feedback signals that improve public-space dominance and functional hierarchy in AI-generated residential floor plans.

  25. A Task-Centric Theory for Iterative Self-Improvement with Easy-to-Hard Curricula

    cs.LG 2026-02 conditional novelty 6.0 of 10

    Iterative self-improvement provably keeps improving only when initial performance lies in a moderate difficulty interval, and easy-to-hard curricula beat fixed mixtures under moderate difficulty separation and suffici...

  26. rePIRL: Learn PRM with Inverse RL for LLM Reasoning

    cs.LG 2026-02 unverdicted novelty 6.0 of 10

    rePIRL learns effective process reward models for LLM reasoning via a dual policy-PRM update process inspired by inverse RL, unifying online and offline methods with reported gains over prior approaches on math and co...

  27. Bridging the Capability Gap: Joint Alignment Tuning for Harmonizing LLM-based Multi-Agent Systems

    cs.CL 2025-09 conditional novelty 6.0 of 10

    MOAT alternately aligns a planning and a grounding LLM agent via perplexity-based DPO and self-generated SFT, reporting small but consistent gains over independently tuned baselines.

  28. Learn from What We HAVE: History-Aware VErifier that Reasons about Past Interactions Online

    cs.RO 2025-08 conditional novelty 6.0 of 10

    A history-aware verifier that scores candidate actions using past interactions cuts failure rates in ambiguous robot manipulation tasks compared to using the generator alone.

  29. Utilizing Training Data to Improve LLM Reasoning for Tabular Understanding

    cs.LG 2025-08 conditional novelty 6.0 of 10

    LRTab retrieves error-avoiding prompt conditions learned from incorrect chain-of-thought traces on training tables to improve LLM tabular reasoning, achieving modest gains on WikiTQ and TabFact.

  30. OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles

    cs.CV 2025-03 conditional novelty 6.0 of 10

    Iterative SFT-RL cycles enable a 7B LVLM to develop sophisticated visual chain-of-thought reasoning and improve performance on math and general reasoning benchmarks.

  31. Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning

    cs.LG 2024-10 unverdicted novelty 6.0 of 10

    Process advantage verifiers trained to predict step-level progress under a distinct prover policy improve LLM reasoning accuracy by over 8% and sample efficiency by 5-6x over outcome reward models.

  32. Training Language Models to Self-Correct via Reinforcement Learning

    cs.LG 2024-09 unverdicted novelty 6.0 of 10

    SCoRe uses multi-turn online RL with regularization on self-generated traces to improve LLM self-correction, achieving 15.6% and 9.1% gains on MATH and HumanEval for Gemini models.

  33. Multi-Turn On-Policy Distillation with Prefix Replay

    cs.LG 2026-07 conditional novelty 5.0 of 10

    By replaying teacher prefixes with a step-decay schedule, multi-turn on-policy distillation can run without live environment interaction, matching or slightly beating online OPD accuracy.

  34. Data Selection Through Iterative Self-Filtering for Vision-Language Settings

    cs.CV 2026-06 unverdicted novelty 5.0 of 10

    An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.

  35. Peak-Then-Collapse and the Four Interface Channels of Knowledge-Graph Tool Use

    cs.CL 2026-05 unverdicted novelty 5.0 of 10

    RL on a four-verb Freebase KG tool shows peak-then-collapse in answer rate, with self-distillation reaching 40% EM while model scaling does not help.

  36. RISE: Reliable Improvement in Self-Evolving Vision-Language Models

    cs.CV 2026-05 unverdicted novelty 5.0 of 10

    RISE proposes a self-evolving VLM framework with three designs to address challenges in question generation and solver adaptation, reporting consistent gains on seven benchmarks across two backbones.

  37. SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation

    cs.AI 2026-03 unverdicted novelty 5.0 of 10

    SOLAR introduces a self-optimizing agent using meta-learning on model weights and RL-driven strategy discovery for lifelong adaptation in LLMs, claiming superior performance on reasoning tasks across domains.

  38. ReST-RL: Achieving Accurate Code Reasoning of LLMs with Optimized Self-Training and Decoding

    cs.AI 2025-08 conditional novelty 5.0 of 10

    ReST-RL pairs reward-variance-filtered self-training (ReST-GRPO) with value-model-guided MCTS decoding (VM-MCTS) to improve LLM code reasoning.

  39. An Iterative Utility Judgment Framework Inspired by Philosophical Relevance via LLMs

    cs.IR 2024-06 unverdicted novelty 5.0 of 10

    ITEM is a new iterative utility judgment loop for RAG that maps Schutz's three levels of relevance to retrieval, utility scoring, and generation, yielding measured gains on TREC DL, WebAP, GTI-NQ, and NQ.

  40. CoTu at EXACT 2026: Neuro-Symbolic Reasoning for Transparent Educational QA

    cs.CL 2026-07 conditional novelty 4.0 of 10

    A 4B LLM that generates solver-runnable code (Z3 for logic, Python for physics) achieved a perfect physics score and the highest technical score (13.44/15) at EXACT 2026, with premise selection as the main remaining e...

  41. Robust Reasoning via Dynamic Token Selection for Distribution-Aligned Self-Distillation

    cs.CL 2026-05 unverdicted novelty 4.0 of 10

    DASD dynamically selects tokens in self-distillation to keep logical corrections while suppressing stylistic noise, improving robustness on math, code, and commonsense benchmarks.

  42. $S^3$-R1: Learning to Retrieve and Answer Step-by-Step with Synthetic Data

    cs.LG 2026-05 unverdicted novelty 4.0 of 10

    S^3-R1 generates synthetic multi-hop questions and uses combined intermediate and final rewards to train RL models for retrieval and answering, reporting up to 10% better out-of-domain generalization.

  43. PRISMA: Preference-Reinforced Self-Training Approach for Interpretable Emotionally Intelligent Negotiation Dialogues

    cs.CL 2026-04 unverdicted novelty 4.0 of 10

    PRISMA augments self-training with direct preference optimization and an emotion-aware negotiation strategy chain-of-thought to produce more interpretable and effective negotiation dialogues on two new datasets.

  44. rePIRL: Learn PRM with Inverse RL for LLM Reasoning

    cs.LG 2026-02 conditional novelty 4.0 of 10

    rePIRL learns token-level process rewards for LLM reasoning via a guided-cost-learning-style IRL objective, and shows these rewards improve reasoning policies on math/coding benchmarks.

  45. Demystifying Long Chain-of-Thought Reasoning in LLMs

    cs.CL 2025-02 unverdicted novelty 4.0 of 10

    Experiments show that long CoT reasoning in LLMs emerges with more training compute when reward shaping is used properly, and scaling verifiable rewards from noisy data helps especially on out-of-distribution tasks.

  46. From System 1 to System 2: A Survey of Reasoning Large Language Models

    cs.AI 2025-02 accept novelty 3.0 of 10

    The survey organizes the shift of LLMs toward deliberate System 2 reasoning, covering model construction techniques, performance on math and coding benchmarks, and future research directions.

  47. LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods

    cs.CL 2024-12 accept novelty 3.0 of 10

    A survey that organizes LLMs-as-judges research into functionality, methodology, applications, meta-evaluation, and limitations.

  48. Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems

    cs.AI 2025-03 unverdicted novelty 2.0 of 10

    This survey frames foundation agents using brain-inspired modular architectures and reviews challenges in evolution, collaboration, and safety.

Pith tools