Pith. sign in

REVIEW 28 cited by

LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!

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 2502.07374 v2 pith:KHQV23NU submitted 2025-02-11 cs.AI

LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!

classification cs.AI
keywords longreasoningmodeltrainingsamplescontentaccuracyachieves
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Large reasoning models (LRMs) tackle complex reasoning problems by following long chain-of-thoughts (Long CoT) that incorporate reflection, backtracking, and self-validation. However, the training techniques and data requirements to elicit Long CoT remain poorly understood. In this work, we find that a Large Language model (LLM) can effectively learn Long CoT reasoning through data-efficient supervised fine-tuning (SFT) and parameter-efficient low-rank adaptation (LoRA). With just 17k long CoT training samples, the Qwen2.5-32B-Instruct model achieves significant improvements on a wide range of math and coding benchmarks, including 56.7% (+40.0%) on AIME 2024 and 57.0% (+8.1%) on LiveCodeBench, competitive to the proprietary o1-preview model's score of 44.6% and 59.1%. More importantly, we find that the structure of Long CoT is critical to the learning process, whereas the content of individual reasoning steps has minimal impact. Perturbations affecting content, such as training on incorrect samples or removing reasoning keywords, have little impact on performance. In contrast, structural modifications that disrupt logical consistency in the Long CoT, such as shuffling or deleting reasoning steps, significantly degrade accuracy. For example, a model trained on Long CoT samples with incorrect answers still achieves only 3.2% lower accuracy compared to training with fully correct samples. These insights deepen our understanding of how to elicit reasoning capabilities in LLMs and highlight key considerations for efficiently training the next generation of reasoning models. This is the academic paper of our previous released Sky-T1-32B-Preview model. Codes are available at https://github.com/NovaSky-AI/SkyThought.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 28 Pith papers

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

  1. Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use

    cs.AI 2026-07 unverdicted novelty 7.0

    Static SFT and RL training for tool-use agents leads to performance drops under open-world distributional shifts across perception, interaction, reasoning and internalization; perturbation-augmented fine-tuning is pro...

  2. LinTree: Improving LLM Reasoning with Explicitly Structured Search Histories

    cs.AI 2026-05 unverdicted novelty 7.0

    Adding explicit parent pointers to represent search tree structure in LLM reasoning traces (LinTree) improves task performance and search efficiency on Blocks World, grid Navigation, and Sokoban relative to implicit t...

  3. Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning

    cs.CL 2026-04 accept novelty 7.0

    Cross-modal agreement between chain-of-thought and program-of-thought reasoning enables self-consistency with only two LLM samples, reducing sampling cost by 9.3x while improving accuracy.

  4. Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning

    cs.CL 2026-04 unverdicted novelty 7.0

    CoT-PoT ensembling achieves self-consistency accuracy in LLMs with only two samples for 78.6% of tasks, reducing computation by 9.3x compared to standard methods.

  5. Embarrassingly Simple Self-Distillation Improves Code Generation

    cs.CL 2026-04 conditional novelty 7.0

    Simple self-distillation—fine-tuning a code model on its own temperature-sampled, truncated outputs—raises LiveCodeBench pass@1 substantially without verifiers, teachers, or RL.

  6. LA-RL: Label-Aware Self-Reflection for Reinforcement Learning in Information Extraction

    cs.CL 2026-07 conditional novelty 6.0

    Label-aware diagnostic reflection plus two-stage outcome GRPO improves same-backbone IE F1 over SFT, with larger gains under relation-extraction domain shift.

  7. The Tell-Tale Norm: $\ell_2$ Magnitude as a Signal for Reasoning Dynamics in Large Language Models

    cs.CL 2026-06 unverdicted novelty 6.0

    The L2 norm of LLM hidden states signals reasoning intensity, with a theoretical bound on SAE feature activations, enabling three new test-time scaling techniques that boost performance.

  8. Quantized Reasoning Models Think They Need to Think Longer, but They Do Not

    cs.LG 2026-05 unverdicted novelty 6.0

    Post-training quantization increases overthinking errors in reasoning models; a logit penalty on curated overthinking markers reduces CoT length 12-23% without accuracy loss.

  9. Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment

    cs.CL 2026-01 conditional novelty 6.0

    Rank-Surprisal Ratio (RSR) correlates strongly (average Spearman 0.86) with post-distillation reasoning gains across five student models and trajectories from eleven teachers, outperforming existing selection metrics.

  10. MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation

    cs.CL 2025-10 conditional novelty 6.0

    A teacher-guided dense reward for GRPO distillation improves small-model tool-use accuracy and out-of-domain generalization on executable-tool benchmarks.

  11. Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimization

    cs.CL 2025-10 conditional novelty 6.0

    LLM attention maps reveal a preplan-and-anchor pattern, and reweighting RL credit toward the flagged tokens improves math/QA reasoning.

  12. Performative Thinking? The Brittle Correlation Between CoT Length and Problem Complexity

    cs.AI 2025-09 conditional novelty 6.0

    A transformer trained to imitate A* produces traces whose length reflects similarity to training data, not true problem complexity, so long chain-of-thought should not be read as more 'thinking'.

  13. Stabilizing Knowledge, Promoting Reasoning: Dual-Token Constraints for RLVR

    cs.CL 2025-07 unverdicted novelty 6.0

    Archer introduces response-level entropy normalization and differentiated clipping/KL regularization in RLVR to encourage exploration on reasoning tokens while stabilizing knowledge tokens, yielding gains in pass@1 an...

  14. LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning

    cs.CL 2025-06 unverdicted novelty 6.0

    LongWriter-Zero applies RL from a base model with specialized rewards for length, quality, and structure to outperform SFT baselines and larger models on long-writing benchmarks.

  15. The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity

    cs.AI 2025-06 unverdicted novelty 6.0

    LRMs exhibit complete accuracy collapse beyond certain puzzle complexities, with reasoning effort rising then declining, outperforming standard LLMs only on medium-complexity tasks.

  16. Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning

    cs.CL 2025-06 conditional novelty 6.0

    High-entropy minority tokens drive RLVR gains, so restricting gradients to the top 20% maintains or improves performance over full updates on Qwen3 models, especially larger ones.

  17. Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces

    cs.CL 2026-06 unverdicted novelty 5.0

    Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.

  18. Absurd World: A Simple Yet Powerful Method to Absurdify the Real-world for Probing LLM Reasoning Capabilities

    cs.AI 2026-05 unverdicted novelty 5.0

    Absurd World automatically converts real-world problems into absurd yet logically coherent scenarios to test whether LLMs can reason without depending on familiar patterns.

  19. How Well Do LLMs Perform on the Simplest Long-Chain Reasoning Tasks: An Empirical Study on the Equivalence Class Problem

    cs.AI 2026-05 unverdicted novelty 5.0

    Non-reasoning LLMs fail the equivalence class problem while reasoning LLMs perform better but remain incomplete, with difficulty peaking at phase transition for the former and maximum diameter for the latter.

  20. Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment

    cs.CL 2026-01 conditional novelty 5.0

    A trajectory for student-LLM distillation is better when its tokens are surprising but still high-ranked, and the ratio of average rank to average surprisal (RSR) selects such trajectories better than existing metrics.

  21. A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning

    cs.LG 2025-10 conditional novelty 5.0

    Reward-free online SFT on a model's own low-temperature samples improves math reasoning and roughly matches GRPO on six benchmarks.

  22. A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning

    cs.LG 2025-10 unverdicted novelty 5.0

    SePT alternates self-generation of responses at controlled temperatures with training on the latest model outputs, yielding gains over a strong no-training baseline on six math reasoning benchmarks.

  23. A Model Can Help Itself: Reward-Free Self-Training for LLM Reasoning

    cs.LG 2025-10 unverdicted novelty 5.0

    SePT enables LLMs to improve math reasoning on multiple benchmarks by iteratively training on their own low-temperature generated responses using an online data refresh mechanism.

  24. Rethinking the Chain-of-Thought: The Roles of In-Context Learning and Pre-trained Priors

    cs.CL 2025-09 conditional novelty 5.0

    The paper finds that as exemplar count increases, LLMs shift from pretrained priors to in-context signals, and long chain-of-thought prompts induce longer reasoning chains that often improve accuracy.

  25. The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes

    cs.CL 2026-06 unverdicted novelty 4.0

    A literature survey that introduces a taxonomy for LLM reasoning paradigms, analyzes methodological trends, and synthesizes failure modes from over 300 papers.

  26. OpenCodeReasoning: Advancing Data Distillation for Competitive Coding

    cs.CL 2025-04 unverdicted novelty 4.0

    A new open SFT dataset for reasoning distillation lets coding models hit state-of-the-art scores on LiveCodeBench and CodeContests with supervised fine-tuning alone, outperforming RL-trained baselines.

  27. ReasoningLens: Hierarchical Visualization and Diagnostic Auditing for Large Reasoning Models

    cs.CL 2026-06 unverdicted novelty 3.0

    ReasoningLens is a framework for hierarchical visualization, agentic auditing, and systemic profiling of long reasoning chains in large reasoning models.

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

    cs.AI 2025-03 unverdicted novelty 2.0

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