Pith. sign in

REVIEW 5 major objections 5 minor 13 cited by

Are Reasoning Models More Prone to Hallucination?

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Whether a reasoning model hallucinates is set by its post-training recipe: cold-start SFT plus verifiable-reward RL reduces hallucination; RL-only and SFT-only increase it.

desk verdict Useful three-way comparison but the SFT+RL takeaway rests on one clean baseline; the behavioral and probing analyses are suggestive, not conclusive. read the letter →

arxiv 2505.23646 v1 pith:4PHKWALR submitted 2025-05-29 cs.CL cs.LG

Zijun Yao , Yantao Liu , Yanxu Chen , Jianhui Chen , Junfeng Fang , Lei Hou , Juanzi Li , Tat-Seng Chua This is my paper · ORCID
classification cs.CLcs.LG
keywords largereasoningmodelshallucinationchain-of-thoughtpost-trainingreinforcementlearningsupervisedfine-tuningmodelcalibrationfactuality
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether large reasoning models—models that think through long chains of reasoning before answering—are more prone to making up facts. It finds the answer depends on how the model was post-trained, not on reasoning ability alone. Models built with a cold-start supervised fine-tuning stage followed by reinforcement learning with verifiable rewards consistently answered more fact-seeking questions correctly than their non-reasoning backbones. Models built by reinforcement learning alone or by distillation alone were generally worse, often repeating the same flawed logic or producing final answers that contradict their own reasoning. The authors conclude that hallucination in reasoning models is largely a training-pipeline effect, and they call for factuality to be tracked during LRM development.

What carries the argument

The argument is carried by a post-training pipeline taxonomy that sorts reasoning models into three recipes: cold-start SFT plus verifiable-reward RL, zero-style RL without cold start, and distillation-only SFT. Each tested model is assigned to one of these categories and compared with its closest non-reasoning backbone on SimpleQA and TriviaQA. The behavioral analysis adds two named failure modes, Flaw Repetition and Think-Answer Mismatch, scored by an LLM judge on wrong outputs. The mechanism analysis then measures expected calibration error and trains a probe on the hidden states of the last question token to test whether correctness information survives post-training.

What would settle it

A controlled experiment using one base model, for example Qwen2.5-7B-Base, post-trained three ways—cold-start SFT then verifiable-reward RL, zero-style RL only, and distillation SFT only—with matched data and compute would settle the claim if the RL-only and SFT-only versions no longer fall below the base on SimpleQA and TriviaQA while the SFT+RL version improves. A cheaper check is to compare DeepSeek-R1 against DeepSeek-V3-Base rather than DeepSeek-V3-Instruct once the base is accessible; if the +4.7 SimpleQA gain disappears, that example no longer supports the SFT+RL takeaway.

Watch

Extended reading notes

Core claim

The paper's central claim is that the post-training pipeline determines a reasoning model's tendency to hallucinate on fact-seeking questions. Across SimpleQA and TriviaQA, all tested SFT+RL LRMs—DeepSeek-R1, Qwen3-32B, GLM-4-Z1-9B, and GLM-4-Z1-32B—outperformed their non-reasoning counterparts, with gains up to +4.7 points on SimpleQA and +4.0 points on TriviaQA. In contrast, RL-only models such as MiMo-7B-RL-Zero, DeepMath-Zero, and DAPO-Qwen-32B mostly lost accuracy, with drops as large as -29.9 points on TriviaQA, and SFT-only distilled models such as the DeepSeek-R1-Distill series and Qwen3-14B also dropped on average. The authors trace the extra errors to two behaviors: flaw repetition, where the model cycles through surface-different but logically identical wrong paths, and think-answer mismatch, where the final answer diverges from the chain-of-thought. They further show the failure is visible as corrupted calibration: RL-only and SFT-only models have higher expected calibration error and lose answer-correctness information in their hidden states, whereas SFT+RL models are better calibrated and their hidden states remain more probeable for correctness.

Load-bearing premise

The load-bearing premise is that the accuracy gaps can be blamed on the post-training pipeline, but the comparisons are not fully controlled: DeepSeek-R1 is compared with DeepSeek-V3-Instruct, GLM-4-Z1-9B with GLM-4-9B-Instruct, and the authors state they could not run the training themselves, so unmeasured differences in base models, instruction tuning, or training schedules could explain part of the effect.

Editorial extensions

If this is right

  • Developers of reasoning models should not assume that long chain-of-thought reasoning improves factuality; fact-seeking accuracy should be evaluated after each post-training stage.
  • RL-only and SFT-only pipelines should carry extra factuality safeguards if used, since models trained this way hallucinate more than their non-reasoning backbones.
  • Calibration error can serve as a practical monitoring signal for hallucination during LRM development, complementing benchmark accuracy.
  • The finding that parameter-update magnitude does not predict hallucination suggests that hallucination is not simply catastrophic forgetting from large weight shifts; the training objective and data matter more.
  • Factuality benchmarks such as SimpleQA and TriviaQA should become standard evaluation companions to formal reasoning tasks when releasing reasoning models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper itself flags that it could not reproduce full-parameter post-training and relies on publicly released models, so the pipeline attribution is an inference across pre-existing models rather than a controlled experiment; a strict test would train one base model through all three pipelines with matched data and compute.
  • Flaw Repetition and Think-Answer Mismatch could be turned into inexpensive real-time hallucination alarms: a decoder that detects semantically repetitive thinking loops or a final answer that contradicts the chain-of-thought could flag high-risk outputs before they are delivered.
  • Because SFT+RL models retained answer-correctness information in hidden states while RL-only and SFT-only models lost it, uncertainty probing could become a model-selection criterion during post-training, separate from benchmark accuracy.
  • Several comparisons use baselines that are not the exact initialization weights, such as DeepSeek-R1 versus DeepSeek-V3-Instruct and GLM-4-Z1-9B versus GLM-4-9B-Instruct; if unmeasured differences in base-model quality or instruction tuning drive part of the deltas, the practical recommendation may become 'evaluate factuality after every recipe' rather than 'prefer SFT+RL unconditionally'.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper asks whether large reasoning models (LRMs) are more prone to hallucination than their non-reasoning counterparts, and answers by comparing factuality on SimpleQA and TriviaQA across models grouped by post-training pipeline. It reports three central takeaways: SFT+RL LRMs improve factuality, RL-only LRMs degrade it, and SFT-only LRMs also degrade it (Section 3.2, Tables 1-3). The paper then introduces two behavioral categories, Flaw Repetition and Think-Answer Mismatch (Section 4), and an uncertainty analysis using self-consistency confidence, ECE calibration, and hidden-state probing (Section 5). It also includes a parameter-shift analysis to argue that catastrophic forgetting is not the main driver (Section 6, Table 7).

Significance. The paper addresses a timely and contested question, releasing evaluation code and covering several model families. Its central direction is plausible: the post-training recipe, not reasoning capability per se, may determine factual reliability. The uncertainty-probing analysis is a distinctive contribution, and the three pipeline-level takeaways are falsifiable. However, the evidence does not yet support the categorical causal generalization in Takeaways #1-3, primarily because the SFT+RL comparison set contains only one initialization-controlled pair, and because most quantitative differences are reported without uncertainty estimates. The paper itself concedes in Section 6 and Appendix A that pretraining corpora, instruction-tuning data, and post-training schedules are uncontrolled.

major comments (5)
  1. [Section 3.2, Table 1] The SFT+RL takeaway is underdetermined by the comparisons. Only GLM-4-Z1-32B vs GLM-4-32B-Base is a true initialization-controlled test. DeepSeek-R1 is compared against DeepSeek-V3-Instruct rather than its actual initialization (V3-Base); GLM-4-Z1-9B is compared against GLM-4-9B-Instruct because the base is unreleased; and the Qwen3-32B row compares the same checkpoint with and without the thinking template, which isolates inference-time reasoning mode rather than the post-training recipe. The Qwen3 comparison is further confounded because the two modes use different decoding parameters (temperature 0.6 vs 0.7, top-p 0.95 vs 0.8; Table 9). Because instruction-tuned baselines can differ from base checkpoints on factuality tasks, the positive deltas in the uncontrolled rows (4.7/2.2 and 1.0/2.4) do not establish that the SFT+RL recipe improves factuality. The paper acknowledges this limitation in Appendix A, but the abstract and Section 3.2 state the takeaway categorically. Please either add controlled model pairs or downgrade the claim to a conditional statement.
  2. [Section 4.1, Table 4] The behavioral analysis uses one representative model per pipeline (GLM-4-Z1-32B, DeepMath-Zero, DPSK-Qwen-32B), and the annotation is performed by Qwen3-32B, the same model family used for answer extraction and correctness grading. The claim that flaw repetition and think-answer mismatch are 'important causes' of hallucination is too strong with n=1 per pipeline, no inter-annotator agreement, and statistics computed only on hallucinated outputs rather than contrasted with correct outputs. In addition, the assertion that non-reasoning counterparts produce no such behaviors may be tautological if those models do not generate a long CoT. Please add more models per pipeline or soften the causal language.
  3. [Section 5.1, Figure 2] The calibration results are reported without error bars or significance tests. The ECE differences are small (e.g., 0.121 vs 0.146 for GLM-Z1-32B vs its base; 0.127 vs 0.112 for DPSK-Qwen-14B vs Qwen2.5-14B-Base), and P(a) is estimated with only N=10 samples per question. It is unclear whether these deltas are within sampling noise. Bootstrap confidence intervals or per-question variance estimates are needed before concluding that SFT+RL improves calibration while SFT-only and RL-only corrupt it.
  4. [Section 3.2, Tables 2 and 3] Several accuracy deltas in the RL-only and SFT-only tables are very small and likely within noise: DeepMath-Zero improves by 0.6 points on SimpleQA, Qwen3-14B drops by only 0.4 on average, and DPSK-Qwen-14B improves by 0.1 on SimpleQA. Without error bars or repeated runs, these deltas cannot support the categorical claim that these pipelines 'are more prone to hallucination.' The paper should report variance estimates or restrict the claims to the larger, consistent effects.
  5. [Section 6, Table 7] The parameter-shift analysis compares raw MAE values across models of different sizes and architectures (Qwen2.5-7B vs Qwen3-14B vs GLM-32B), so the absolute values are not directly comparable. The observation that GLM-Z1-32B has the largest MAE yet mild hallucination, while smaller-MAE models hallucinate more, could be an artifact of scale or architecture differences rather than evidence against catastrophic forgetting. A normalized or within-family comparison would be more convincing.
minor comments (5)
  1. [Table 4] The header contains a typo: 'Fraw Rep.' should be 'Flaw Rep.'
  2. [Section 4.1, Table 4] The text refers to 'DPSK-Qwen-7B' in the behavior discussion, but the model collection in Section 3.2 lists DPSK-Qwen-14B, DPSK-Qwen-32B, DPSK-LLaMA-8B, and DPSK-LLaMA-70B; the 7B name appears only in Table 4 and should be resolved.
  3. [Appendix E] There is a typo: 'verision 0.8.3' should be 'version 0.8.3'.
  4. [Section 3.1] The text says 'we omit the provided excepts' — 'excepts' should be 'excerpts'.
  5. [Table 3] The table only uses the ↓ arrow, but the text notes that DPSK-LLaMA-70B improves on TriviaQA (74.1 to 75.2); please mark this improvement with ↑ or explain the convention in the caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the accuracy claims rest on external benchmarks, and the baseline and judge concerns are methodological confounds rather than construction-identity circularity.

full rationale

The paper's central claim—that post-training pipeline (SFT+RL vs RL-only vs SFT-only) affects hallucination tendency—is supported by accuracy measurements on SimpleQA and TriviaQA, which are external benchmarks with gold answers. No parameter is fitted to those accuracies and then renamed as a prediction; the takeaways in Section 3.2 simply summarize measured deltas. The SFT+RL conclusion is an empirical generalization from released model families, not a quantity derived from its own assumptions. The skeptic's points about uncontrolled baselines (DeepSeek-R1 vs DeepSeek-V3-Instruct, GLM-4-Z1-9B vs GLM-4-9B-Instruct, Qwen3 self-comparison) are threats to causal attribution and internal validity, but they do not make any result true by definition or by construction: the backbones are external checkpoints, not outputs of the paper's own equations. The use of Qwen3-32B as judge is a shared measurement instrument and could bias comparisons, but it is not the same variable as the hallucination rate being claimed, so this is a confound rather than circularity. The paper itself acknowledges the lack of strict variable control in Section 6 and Appendix A, which is a limitation, not a circular step. Self-citations (e.g., [13] for GLM-4-Z1) identify the evaluated models and are not load-bearing for the derivation: the measured accuracies stand independently of those citations. No equation in the paper reduces a claimed result to an input by construction, and no prediction is statistically forced by a fitted parameter. Accordingly, the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper rests on a set of domain assumptions about benchmark validity, model-pipeline taxonomy, confidence proxies, and judge neutrality. These are not derived in the paper; they are standard practice but are load-bearing for the empirical conclusions.

assumptions (5)
  • domain assumption SimpleQA and TriviaQA, evaluated by LLM-as-a-Judge using the SimpleQA grader template, validly measure hallucination.
    The paper equates hallucination with factual incorrectness on these benchmarks (Section 3.1) but acknowledges judge subjectivity in the Limitations.
  • domain assumption Publicly released models can be reliably classified into SFT+RL, RL-only, and SFT-only pipelines based on the cited technical reports and blog posts.
    The entire comparison depends on this taxonomy, which the paper uses in Section 3.1 and Tables 1-3.
  • domain assumption The self-consistency probability P(a), estimated by 10 samples and majority voting, is a valid proxy for model confidence.
    Used as the x-axis in the calibration plots (Section 5.1); this is a known proxy but not the model's stated probability.
  • domain assumption Probing the last-token hidden state of the question with a logistic regression can recover the model's internal uncertainty about the answer.
    Section 5.2; the layer and normalization details are not specified, and the below-chance result for MiMo-7B-RL-Zero suggests the premise may fail.
  • domain assumption LLM-as-a-Judge with Qwen3-32B is unbiased across the evaluated model families.
    Used for answer extraction, correctness grading, and behavior annotation (Sections 3.1, 4.1, 5.1); the judge is itself an SFT+RL model, which is the pipeline the paper finds most favorable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Are Reasoning Models More Prone to Hallucination?." pith.science (2026). https://pith.science/paper/4PHKWALR

@misc{pith2026250523646,
  author       = {Pith},
  title        = {Pith review of: Are Reasoning Models More Prone to Hallucination?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4PHKWALR}},
  note         = {Machine review of arXiv:2505.23646}
}
read the original abstract

Recently evolved large reasoning models (LRMs) show powerful performance in solving complex tasks with long chain-of-thought (CoT) reasoning capability. As these LRMs are mostly developed by post-training on formal reasoning tasks, whether they generalize the reasoning capability to help reduce hallucination in fact-seeking tasks remains unclear and debated. For instance, DeepSeek-R1 reports increased performance on SimpleQA, a fact-seeking benchmark, while OpenAI-o3 observes even severer hallucination. This discrepancy naturally raises the following research question: Are reasoning models more prone to hallucination? This paper addresses the question from three perspectives. (1) We first conduct a holistic evaluation for the hallucination in LRMs. Our analysis reveals that LRMs undergo a full post-training pipeline with cold start supervised fine-tuning (SFT) and verifiable reward RL generally alleviate their hallucination. In contrast, both distillation alone and RL training without cold start fine-tuning introduce more nuanced hallucinations. (2) To explore why different post-training pipelines alters the impact on hallucination in LRMs, we conduct behavior analysis. We characterize two critical cognitive behaviors that directly affect the factuality of a LRM: Flaw Repetition, where the surface-level reasoning attempts repeatedly follow the same underlying flawed logic, and Think-Answer Mismatch, where the final answer fails to faithfully match the previous CoT process. (3) Further, we investigate the mechanism behind the hallucination of LRMs from the perspective of model uncertainty. We find that increased hallucination of LRMs is usually associated with the misalignment between model uncertainty and factual accuracy. Our work provides an initial understanding of the hallucination in LRMs.

Figures

Figures reproduced from arXiv: 2505.23646 by the authors.

Figure 1
Figure 1. Comparison of factual accuracy between LRMs and their backbone counterparts, along [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Calibration plot comparing LRMs with their non-reasoning counterparts on TriviaQA. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 13 Pith papers

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

  1. Reasoning Denoiser: Denoising Reasoning Traces for Hallucination Detection in Large Reasoning Models

    cs.AI 2026-07 conditional novelty 6.0 of 10

    REDE removes irrelevant and repetitive reasoning steps using final-answer attention as supervision, improving hallucination detection AUROC by up to ~19 points over unfiltered traces.

  2. Different Teachers, Different Capabilities: Sub-1B On-Device Distillation for Structured Text Enrichment

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Distilling an 8B reasoning teacher into a 0.6B student recovers most summary quality at ~50× speed, but teacher type—not scale alone—determines which capabilities transfer.

  3. Mitigating Factual Hallucination in Large Reasoning Models via Mixed-Mode Advantage Regularization

    cs.CL 2026-07 conditional novelty 6.0 of 10

    MARGO mitigates thinking-induced hallucination in large reasoning models by using mixed-mode GRPO rollout groups that compare thinking trajectories against same-model non-thinking references.

  4. UCPO: Uncertainty-Aware Policy Optimization

    cs.AI 2026-01 conditional novelty 6.0 of 10

    UCPO uses decoupled advantage normalization and a dynamic uncertainty reward to train LLMs to express uncertainty, improving answer reliability (PAQ) but sometimes lowering F1.

  5. PBFuzz: Agentic Directed Fuzzing for PoV Generation

    cs.CR 2025-12 conditional novelty 6.0 of 10

    An agentic fuzzing system lets LLM agents infer vulnerability constraints, encode them as parameter generators, and solve them with property-based testing; it triggered 57 Magma CVEs, 17 missed by other fuzzers.

  6. Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models

    cs.LG 2025-10 conditional novelty 6.0 of 10

    A dynamic replay and reweighting scheduler (RECAP) preserves general capabilities during RLVR while keeping reasoning performance at least as good as reasoning-only finetuning.

  7. Locus: Agentic Predicate Synthesis for Directed Fuzzing

    cs.CR 2025-08 conditional novelty 6.0 of 10

    Locus uses an LLM agent to synthesize and validate intermediate predicates that make directed fuzzing reach target bug states faster, reporting an average 41.6x speedup across eight fuzzers.

  8. REIN: Bridging the Gap between Reasoning and Reliability via Reflection and Abstention Alignment

    cs.AI 2026-08 conditional novelty 5.0 of 10

    REIN trains reasoning models with reflection-veracity and abstention rewards, letting them answer or abstain in a single pass, and reports large drops in a false-endorsement metric across four benchmarks.

  9. Reasoning Error from Known Fact: Step-Level Self-Consistency Group Relative Policy Optimization for LLM

    cs.CL 2026-07 conditional novelty 5.0 of 10

    SSC-GRPO reweights GRPO advantages using step-level self-consistency judgments to reduce context-sensitive factual hallucinations, with modest average gains across math and hallucination benchmarks.

  10. Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewards

    cs.LG 2025-09 conditional novelty 5.0 of 10

    The paper identifies confounds in RLVR evaluations that inflate apparent gains and proposes a minimum standard for budget-matched, contamination-aware assessment with calibration tracking.

  11. Mitigating Think-Answer Mismatch in LLM Reasoning Through Noise-Aware Advantage Reweighting

    cs.LG 2025-08 reject novelty 5.0 of 10

    S-GRPO reweights GRPO advantages by an assumed noise level p to down-weight unbalanced groups, claiming improved and more noise-robust math reasoning training.

  12. A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models

    cs.CL 2025-09 conditional novelty 4.0 of 10

    A structured literature survey concluding that reasoning capabilities do not automatically make LLMs more trustworthy and can introduce new vulnerabilities in safety, robustness, and privacy.

  13. Towards Concise and Adaptive Thinking in Large Reasoning Models: A Survey

    cs.AI 2025-07 conditional novelty 3.0 of 10

    A comprehensive review that categorizes methods for shortening and adaptively triggering chain-of-thought reasoning in large language models.

Reference graph

Works this paper leans on

43 extracted references · 27 canonical work pages · cited by 13 Pith papers

  1. [1]

    Benchmarking foundation models with language- model-as-an-examiner

    Yushi Bai, Jiahao Ying, Yixin Cao, Xin Lv, Yuze He, Xiaozhi Wang, Jifan Yu, Kaisheng Zeng, Yijia Xiao, Haozhe Lyu, Jiayin Zhang, Juanzi Li, and Lei Hou. Benchmarking foundation models with language- model-as-an-examiner. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, N...

    work page 2023

  2. [2]

    Probabilistic tree-of-thought reasoning for answering knowledge-intensive complex questions

    Shulin Cao, Jiajie Zhang, Jiaxin Shi, Xin Lv, Zijun Yao, Qi Tian, Lei Hou, and Juanzi Li. Probabilistic tree-of-thought reasoning for answering knowledge-intensive complex questions. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 12541–12560, 2023. URL https: //aclanthology.org/2023.findings-emnlp.835/

    work page 2023

  3. [3]

    Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. Program of thoughts prompting: Disentangling computation from reasoning for numerical reasoning tasks.Trans. Mach. Learn. Res., 2023,

    work page 2023

  4. [4]

    Evaluating generative lan- guage models in information extraction as subjective question correction

    Yuchen Fan, Yantao Liu, Zijun Yao, Jifan Yu, Lei Hou, and Juanzi Li. Evaluating generative lan- guage models in information extraction as subjective question correction. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evalu- ation, LREC/COLING 2024, 20-25 May, 2024, Torino, Italy, pages 6409–64...

    work page 2024

  5. [5]

    Specializing smaller language models towards multi-step reasoning

    Yao Fu, Hao Peng, Litu Ou, Ashish Sabharwal, and Tushar Khot. Specializing smaller language models towards multi-step reasoning. InInternational Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, pages 10421–10430, 2023. URL https://proceedings.mlr.press/ v202/fu23d.html

    work page 2023

  6. [6]

    Strategic reasoning with language models

    Kanishk Gandhi, Dorsa Sadigh, and Noah Goodman. Strategic reasoning with language models. In NeurIPS 2023 Foundation Models for Decision Making Workshop, 2023. URL https://arxiv.org/ pdf/2305.19165

    arXiv 2023

  7. [7]

    PAL: program-aided language models

    Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. PAL: program-aided language models. InInternational Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, pages 10764–10799, 2023. URLhttps://proceedings. mlr.press/v202/gao23f.html

    work page 2023

  8. [8]

    A survey of confidence estimation and calibration in large language models

    Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, and Iryna Gurevych. A survey of confidence estimation and calibration in large language models. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 6577–6595, 2024. UR...

    work page 2024

Show all 43 references
  1. [9]

    Glm-4-0414 model series.https://github.com/THUDM/GLM-4, 2025

    Team GLM. Glm-4-0414 model series.https://github.com/THUDM/GLM-4, 2025

    2025

  2. [10]

    Deliberative alignment: Reasoning enables safer language models.arXiv preprint CoRR, abs/2412.16339, 2024

    Melody Y Guan, Manas Joglekar, Eric Wallace, Saachi Jain, Boaz Barak, Alec Helyar, Rachel Dias, Andrea Vallone, Hongyu Ren, Jason Wei, et al. Deliberative alignment: Reasoning enables safer language models.arXiv preprint CoRR, abs/2412.16339, 2024. URLhttps://arxiv.org/pdf/2412.16339

    2024 arXiv

  3. [11]

    Deepseek-R1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-R1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    2025 arXiv

  4. [12]

    Deepmath-103k: A large-scale, challenging, decontaminated, and verifiable mathematical dataset for advancing reasoning.arXiv preprint CoRR, abs/2504.11456, 2025

    Zhiwei He, Tian Liang, Jiahao Xu, Qiuzhi Liu, Xingyu Chen, Yue Wang, Linfeng Song, Dian Yu, Zhenwen Liang, Wenxuan Wang, et al. Deepmath-103k: A large-scale, challenging, decontaminated, and verifiable mathematical dataset for advancing reasoning.arXiv preprint CoRR, abs/2504....

    2025 arXiv

  5. [13]

    Advancing language model reasoning through reinforcement learning and inference scaling.arXiv preprint CoRR, abs/2501.11651, 2025

    Zhenyu Hou, Xin Lv, Rui Lu, Jiajie Zhang, Yujiang Li, Zijun Yao, Juanzi Li, Jie Tang, and Yuxiao Dong. Advancing language model reasoning through reinforcement learning and inference scaling.arXiv preprint CoRR, abs/2501.11651, 2025. URLhttps://arxiv.org/pdf/2501.11651. 10

    2025 arXiv

  6. [14]

    Openai o1 system card.arXiv preprint CoRR, abs/2412.16720, 2024

    Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, et al. Openai o1 system card.arXiv preprint CoRR, abs/2412.16720, 2024. URLhttps://arxiv.org/pdf/2412.16720

    2024 arXiv

  7. [15]

    A survey on large language model hallucination via a creativity perspective.arXiv preprint CoRR, abs/2402.06647, 2024

    Xuhui Jiang, Yuxing Tian, Fengrui Hua, Chengjin Xu, Yuanzhuo Wang, and Jian Guo. A survey on large language model hallucination via a creativity perspective.arXiv preprint CoRR, abs/2402.06647, 2024. URLhttps://arxiv.org/pdf/2402.06647

    2024 arXiv

  8. [16]

    Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension

    Mandar Joshi, Eunsol Choi, Daniel S Weld, and Luke Zettlemoyer. Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. InProceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16...

    2017

  9. [17]

    Large language models are zero-shot reasoners

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language models are zero-shot reasoners. InAdvances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans,...

    2022

  10. [18]

    T \" ulu 3: Pushing frontiers in open language model post-training.arXiv preprint CoRR, abs/2411.15124, 2024

    Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, et al. T \" ulu 3: Pushing frontiers in open language model post-training.arXiv preprint CoRR, abs/2411.15124, 2024. URL h...

    2024 arXiv

  11. [19]

    Rlaif vs

    Harrison Lee, Samrat Phatale, Hassan Mansoor, Thomas Mesnard, Johan Ferret, Kellie Ren Lu, Colton Bishop, Ethan Hall, Victor Carbune, Abhinav Rastogi, et al. Rlaif vs. rlhf: Scaling reinforcement learning from human feedback with ai feedback. InForty-first International Confer...

    2024

  12. [20]

    Deepseek-v3 technical report.arXiv preprint CoRR, abs/2412.19437, 2024

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. Deepseek-v3 technical report.arXiv preprint CoRR, abs/2412.19437, 2024. URLhttps://arxiv.org/pdf/2412.19437

    2024 arXiv

  13. [21]

    Deepscaler: Surpassing o1-preview with a 1.5 b model by scaling rl.Notion Blog, 2025

    Michael Luo, Sijun Tan, Justin Wong, Xiaoxiang Shi, William Y Tang, Manan Roongta, Colin Cai, Jeffrey Luo, Tianjun Zhang, Li Erran Li, et al. Deepscaler: Surpassing o1-preview with a 1.5 b model by scaling rl.Notion Blog, 2025

    2025

  14. [23]

    Teaching small language models to reason

    Lucie Charlotte Magister, Jonathan Mallinson, Jakub Adamek, Eric Malmi, and Aliaksei Severyn. Teaching small language models to reason. InProceedings of the 61st Annual Meeting of the As- sociation for Computational Linguistics (Volume 2: Short Papers), pages 1773–1781, 2023. ...

    2023

  15. [24]

    URLhttps://arxiv.org/pdf/2308.08747

    arXiv

  16. [25]

    Openai o3 and o4-mini system card

    OpenAI. Openai o3 and o4-mini system card. 2025. Published April 16, 2025

    2025

  17. [26]

    Cooper, and Milos Hauskrecht

    Mahdi Pakdaman Naeini, Gregory F. Cooper, and Milos Hauskrecht. Obtaining well calibrated probabilities using bayesian binning. InProceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence, January 25-30, 2015, Austin, Texas, USA, pages 2901–2907, 2015. URL htt...

    2015

  18. [27]

    Recursive introspection: Teaching language model agents how to self-improve

    Yuxiao Qu, Tianjun Zhang, Naman Garg, and Aviral Kumar. Recursive introspection: Teaching language model agents how to self-improve. InAdvances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver...

    2024

  19. [28]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kel- ton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Le...

    2022

  20. [29]

    A comprehensive survey of hallucination in large language, image, video and audio foundation models

    Pranab Sahoo, Prabhash Meharia, Akash Ghosh, Sriparna Saha, Vinija Jain, and Aman Chadha. A comprehensive survey of hallucination in large language, image, video and audio foundation models. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 11709–...

    2024

  21. [30]

    Qwen3: Think deeper, act faster

    Team Qwen. Qwen3: Think deeper, act faster. https://qwenlm.github.io/blog/qwen3, 2025. Accessed: 2025-04-29

    2025

  22. [31]

    MiMo: Unlocking the reasoning potential of language model – from pretraining to posttraining.arXiv preprint CoRR, abs/2505.07608, 2025

    Xiaomi LLM-Core Team. MiMo: Unlocking the reasoning potential of language model – from pretraining to posttraining.arXiv preprint CoRR, abs/2505.07608, 2025. URL https://arxiv.org/pdf/2505. 07608

    2025 arXiv

  23. [32]

    Mastering the game of go without human knowledge.Nat., 550:354–359, 2017

    David Silver, Julian Schrittwieser, Karen Simonyan, Ioannis Antonoglou, Aja Huang, Arthur Guez, Thomas Hubert, Lucas Baker, Matthew Lai, Adrian Bolton, et al. Mastering the game of go without human knowledge.Nat., 550:354–359, 2017. URLhttps://www.nature.com/articles/nature24270

    2017

  24. [33]

    Chi, Quoc V

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V . Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language mod- els. InAdvances in Neural Information Processing Systems 35: Annual Conference on Neural I...

    2022

  25. [34]

    ReFT: Reasoning with reinforced fine-tuning

    Luong Trung, Xinbo Zhang, Zhanming Jie, Peng Sun, Xiaoran Jin, and Hang Li. ReFT: Reasoning with reinforced fine-tuning. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7601–7614, 2024. URL https://aclant...

    2024

  26. [35]

    Towards large reasoning models: A survey of reinforced reasoning with large language models.arXiv preprint CoRR, abs/2501.09686, 2025

    Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, et al. Towards large reasoning models: A survey of reinforced reasoning with large language models.arXiv preprint CoRR, abs/2501.09686, 2025. ...

    2025 arXiv

  27. [36]

    Measuring short-form factuality in large language models.arXiv preprint CoRR, abs/2411.04368, 2024

    Jason Wei, Nguyen Karina, Hyung Won Chung, Yunxin Joy Jiao, Spencer Papay, Amelia Glaese, John Schulman, and William Fedus. Measuring short-form factuality in large language models.arXiv preprint CoRR, abs/2411.04368, 2024. URLhttps://arxiv.org/pdf/2411.04368

    2024 arXiv

  28. [37]

    LIMO: Less is more for reasoning.arXiv preprint CoRR, abs/2502.03387, 2025

    Yixin Ye, Zhen Huang, Yang Xiao, Ethan Chern, Shijie Xia, and Pengfei Liu. LIMO: Less is more for reasoning.arXiv preprint CoRR, abs/2502.03387, 2025. URLhttps://arxiv.org/pdf/2502.03387

    2025 arXiv

  29. [38]

    Qwen3 technical report.arXiv preprint CoRR, abs/2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, et al. Qwen3 technical report.arXiv preprint CoRR, abs/2505.09388, 2025. URLhttps://arxiv.org/pdf/2505.09388

    2025 arXiv

  30. [39]

    STaR: Bootstrapping reasoning with reason- ing

    Eric Zelikman, Yuhuai Wu, Jesse Mu, and Noah Goodman. STaR: Bootstrapping reasoning with reason- ing. pages 15476–15488, 2022. URL http://papers.nips.cc/paper_files/paper/2022/hash/ 639a9a172c044fbb64175b5fad42e9a5-Abstract-Conference.html

    2022

  31. [40]

    DAPO:an open-source llm reinforcement learning system at scale.arXiv preprint CoRR, abs/2503.14476, 2025

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Tiantian Fan, Gaohong Liu, Lingjun Liu, Xin Liu, et al. DAPO:an open-source llm reinforcement learning system at scale.arXiv preprint CoRR, abs/2503.14476, 2025. URLhttps://arxiv.org/pdf/2503.14476

    2025 arXiv

  32. [41]

    Xing, Hao Zhang, Joseph E

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric P. Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. Judging llm-as-a-judge with mt-bench and chatbot arena. InAdvances in Neural Information Processi...

    2023

  33. [42]

    Automatic chain of thought prompting in large language models

    Zhuosheng Zhang, Aston Zhang, Mu Li, and Alex Smola. Automatic chain of thought prompting in large language models. InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023, 2023. URLhttps://openreview.net/forum?id=5NTt8GFjUHkr

    2023

  34. [44]

    intelligent

    Chiwei Zhu, Benfeng Xu, Quan Wang, Yongdong Zhang, and Zhendong Mao. On the calibration of large language models and alignment. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 9778–9795, 2023. URLhttps://aclanthology.org/2023.findings-emnlp.654/....

    2023

  35. [2023]

    URLhttps://openreview.net/forum?id=YfZ4ZPt8zd

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.