Pith. sign in

REVIEW 5 major objections 5 minor 47 references

Refining Over Resampling: Test-Time Self-Correction for LLM Reasoning

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

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

Pith's one-line read LLM math reasoning improves when sampled answers are critiqued and repaired before majority voting, rather than when more samples or a verifier are used.

desk verdict Genuinely tested breadth-depth refinement, but the title claim is unsupported without an equal-compute baseline. read the letter →

arxiv 2608.05643 v1 pith:UN4EWNYV submitted 2026-08-06 cs.AI cs.CL

Ahsan Bilal , Muhammad Ahmed Mohsin , Muhammad Umer , Lena Trigg , Ali Subhan , Muhammad Ali , Dean F. Hougen This is my paper · ORCID
classification cs.AIcs.CL
keywords test-timescalingself-correctionLLMreasoningmajorityvotingverifier-freemathematicalbreadth-depthrefinementself-critique
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

The paper's central claim is that extra inference compute for LLM reasoning produces more accuracy when spent on repairing the solutions the model already sampled than on simply sampling more solutions or relying on an external reward model. It proposes a training-free, verifier-free breadth–depth refinement method: sample $N$ independent reasoning rollouts, run $D$ rounds of self-critique and self-correction on each rollout using the same base model in three roles — generator, critic, corrector — and aggregate the terminal answers by majority voting. Across AIME24, AIME25, AMC, OlympiadBench, and MATH500, with four open-weight models, the method consistently beats greedy decoding, majority voting, verifier-based best-of-$N$, beam search, and lookahead decoding. The largest reported gain is on the smallest model: Qwen2.5-1.5B rises from 29.6% to 58.0% on MATH500. The paper argues this succeeds because depth repairs local reasoning errors before aggregation, while voting suppresses the noisy, sometimes harmful, flips that occur in individual corrections.

What carries the argument

The central machinery is a three-stage refinement cycle applied for $D$ depths to each of $N$ independent rollouts: reasoning continuation (the generator extends or rewrites the previous trace), self-critique (the critic names logical, arithmetic, or structural errors or explicitly confirms the trace), and self-correction (the corrector revises the trace conditioned on both). The same base model is role-conditioned by prompts, so no external verifier, process reward model, learned stop policy, or additional training is introduced. The cycle's work is to give each trajectory repeated opportunities to repair local errors; the final parameter-free plurality vote then aggregates the refined terminal answers, relying on cross-rollout agreement to suppress the noisy per-rollout corrections documented in Figure 4.

What would settle it

Run the method on AIME24 with Qwen2.5-1.5B at $N=8$, $D=4$, replacing the self-critique prompt with one that always affirms the current trace: if the refined majority-vote accuracy does not fall below unrefined Maj@8, then the claimed benefit is not caused by the critique-and-correct signal, and the loop's value would need another explanation.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that refinement — not more resampling and not verifier selection — is the more effective use of test-time compute for mathematical reasoning. Given a problem $x$, the method draws $N$ initial rollouts from the model's generator policy, then for each of $D$ depths produces a reasoning continuation, a natural-language self-critique, and a corrected trace, all with the same base model and no external reward model; terminal traces are mapped to normalized answers and combined by plurality vote. The paper motivates the design with two failure modes: diversity saturation, where extra samples keep revisiting existing reasoning clusters, and a per-trace hallucination floor, where every i.i.d. rollout is exposed to the same chance of perturbing the intended answer away from correct. Its own diagnostics show that per-rollout correction is noisy and can have negative net benefit, so the final claim is that majority voting over refined rollouts retains the shared repairs and suppresses isolated regressions. The paper expects this to matter most for weaker models, where latent reasoning ability exists but is not reliably expressed in a single first-pass answer.

Load-bearing premise

The load-bearing premise is that the same base model, prompted as critic and corrector, produces enough net-valid corrections that majority voting over refined rollouts beats majority voting over the original rollouts, even though the paper's own Figure 4 shows per-rollout correction can have negative net benefit and its Limitations section concedes that over-correction can introduce regressions.

Editorial extensions

If this is right

  • Because the method needs no reward model or extra training, it can be applied to any open-weight model at inference time and should transfer across domains where answers are extractable and voteable.
  • With $N=8$ and $D=4$ fixed, the reported gains are consistent across five benchmarks and across all compared baselines; weaker models gain most, which suggests refinement recovers latent reasoning ability.
  • The terminal rollout diversity remains high (0.80–0.97 at depth 4), so the voting step is meaningful rather than a rubber stamp of convergent rollouts.
  • Compute-normalized gains are positive against every baseline, with the largest per-TFLOP gains at small model scale; equal-sample comparisons may still understate the method's compute cost.
  • The explicit self-critique stage adds accuracy beyond pure correction in most settings (Table 2), meaning the natural-language critique is information-bearing and not just extra tokens.

Reading between the lines

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

  • Beyond the paper's claims, one can test whether the framework's benefit is really from the critique stage by replacing the critic prompt with an always-confirm prompt; if refined voting accuracy survives that change, the gain would trace to something other than self-critique quality.
  • Beyond the paper's claims, the same repair-before-aggregate recipe should transfer to code generation or formal proof tasks, where a symbolic checker can play the role the plurality vote plays here; this is a natural next benchmark.
  • Beyond the paper's claims, the compute-normalized results suggest an untested scaling hypothesis: for a fixed budget, there is an optimal allocation between $N$ and $D$, and the paper's own Table 3 already shows accuracy is not monotonic in either coordinate.
  • Beyond the paper's claims, the method's dependence on majority voting means it will likely underperform on single-solution tasks or when the model's self-critique is systematically overconfident; a testable variant would weigh refined rollouts by confidence instead of applying a flat vote.
Share X Bluesky LinkedIn Reddit HN

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 proposes a training-free, verifier-free test-time reasoning framework that combines breadth (N independently sampled rollouts) with depth (D rounds of self-critique and self-correction applied to each rollout) and then aggregates the refined rollouts by majority voting. The method is evaluated on AIME24, AIME25, AMC, OlympiadBench, and MATH500 with four open-weight models, reporting accuracy gains over greedy decoding, majority voting, verifier-based best-of-N, beam search, and lookahead decoding, together with TFLOP accounting, compute-normalized gain metrics, ablations of the critic stage, and quantitative diagnostics of refinement dynamics. The paper also includes a limitation section acknowledging the higher inference cost and the dependence on the base model's self-critique ability.

Significance. If the central claim is established, the work would be a useful and timely contribution: it offers a simple, model-agnostic, verifier-free test-time scaling strategy that requires no additional training and no reward model, and it provides a rare transparent discussion of the compute trade-off and of the noisy, sometimes harmful, nature of per-rollout self-correction. The paper includes several strengths: an explicit ablation of the self-critique stage (Table 2), a breadth-depth allocation study (Table 3), compute accounting and compute-normalized gains (Table 4, Figures 5-6), and honest diagnostics showing that per-rollout correction benefit can be negative (Figure 4). However, the main scientific claim that refinement is more effective than resampling is not yet established, because the experiments do not include an equal-compute resampling baseline, the abstract overstates the consistency of the gains relative to Table 1, and the headline numbers lack any measure of statistical uncertainty. These issues are load-bearing for the central claim and require revision.

major comments (5)
  1. [§5.3, Table 4] The central claim that refining sampled trajectories is more effective than resampling requires an equal-compute comparison, but all main results compare methods at a fixed sample count N=8. Under the paper's own TFLOP accounting, Ours uses roughly 728 additional TFLOPs over Maj@8 for Qwen2.5-1.5B and about 2,184-2,624 additional TFLOPs for the 7B/8B models (Table 4). Section 5.3 correctly states that the compute-normalized gain eta 'does not constitute a direct equal-budget comparison in which majority voting is scaled to the same compute.' Without a Maj@K baseline scaled to the same TFLOP budget as D=4 refinement, the observed accuracy advantage could be due to the extra forward passes rather than to the depth of refinement. Please add an equal-compute majority-voting baseline (e.g., Maj@K with K chosen so that total TFLOPs match Ours) and report whether the advantage survives.
  2. [Abstract; Table 1] The abstract's claim that the method 'consistently improves over greedy decoding, majority voting, verifier-based best-of-N, beam search, and lookahead decoding' is contradicted by Table 1. For example, on AIME24 with Qwen2.5-Math-7B, Ours (10.0) is below RM@8 (13.33); on AIME25 with LLaMA-3.1-8B, Ours (6.7) is below Greedy, Maj@8, and RM@8 (all 10.0); and on AIME25 with Ministral-8B, Ours (3.33) ties RM@8 (3.33). Several other cells are ties with Maj@8. The Table 4 W/T/L counts (e.g., 4/1/0 for Qwen2.5-Math-7B vs RM@8, and 2/1/2 for LLaMA-3.1-8B vs RM@8) are more accurate. Please soften the abstract claim to 'generally improves' or 'improves on most benchmark-model pairs' and report the W/T/L summary in the abstract or at least in the results section.
  3. [§4.1, Table 1] The reported accuracies lack error bars, multiple seeds, or significance tests. AIME24 and AIME25 have only 30 questions each, so a 3.33-percentage-point difference is a single question, and several of the claimed gains are exactly one question. For instance, the Qwen2.5-1.5B AIME24 gain from 3.33 (Maj@8/RM@8) to 6.67 (Ours) is one question, and the LLaMA-3.1-8B AIME25 loss from 10.0 to 6.7 is one question. Without multiple seeds or bootstrap confidence intervals, the headline improvements on MATH500 (e.g., 29.6 to 58.0 for Qwen2.5-1.5B) could be partly due to sampling noise. Please report at least three seeds with mean and standard deviation for the main results, or provide bootstrap CIs over questions.
  4. [§5.2, Figure 4] Figure 4 shows that the per-rollout net correction benefit B(d) is negative at several depths for some models (e.g., the Qwen2.5-Math-7B curve dips below zero at early depths), and the paper's own limitation section acknowledges that refinement 'can introduce rollout-level regressions.' The argument that majority voting suppresses these regressions is plausible but is not directly tested. The paper reports vote-level recovery and regression rates in Figure 3 but does not provide a statistical test of whether the vote-level net benefit (recovery minus regression) is significantly positive across questions, nor does it quantify how sensitive the final accuracy is to the per-rollout flip rates. Please add a paired bootstrap or sign test comparing depth-0 vote accuracy with terminal vote accuracy per question, and report the per-question distribution of vote-level net benefit.
  5. [§3.3, Appendix B] The method is entirely defined by the role-conditioned prompts used for the generator, critic, and corrector, yet the paper never provides these prompt templates. Appendix B shows only a single qualitative trace for one problem and one model. Without the exact prompts, the experiments are not reproducible and the claimed generality across models cannot be independently verified. Please include the full prompt templates (with the exact role instructions, output format constraints, and any special tokens) in an appendix or supplementary material.
minor comments (5)
  1. [Table 1] There are several typographical issues in Table 1: 'Ours10.0' is missing a space in the Qwen2.5-Math-7B row, '0.06.67' in the Qwen2.5-1.5B Greedy row should be '0.0 6.67', and the Ministral-8B MATH cell '65.78' appears to be a rendering of '65.78' but should be checked against the value 65.78 in the text.
  2. [Figure 6] The y-axis label of Figure 6 reads 'CNG2[acc.pp/103TFLOPs]' which appears to be a typo for 'CNG' (compute-normalized gain) and the units should be formatted consistently with Table 4 and Appendix A.7.
  3. [Equation (2)] The perturbation model in Eq. (2) uses constants lambda and h from Liu et al. (2024) without checking whether they apply to the models evaluated here. The paper should state clearly that this equation is an illustrative first-order model, not an empirical characterization of the tested models, and that the i.i.d. perturbation assumption may not hold if errors are correlated across samples.
  4. [Appendix A.6] The clustering threshold tau_sim=0.85 and the choice of all-MiniLM-L6-v2 are free parameters of the diversity-saturation analysis in Figure 1. The appendix notes that answer-based clustering is the primary grouping, but the redundancy curves in Figure 1 depend on the sentence-embedding clustering; a sensitivity check over tau_sim (e.g., 0.75, 0.85, 0.95) would strengthen the motivation.
  5. [Algorithm 1] Algorithm 1 says the critic stage is 'parallel / batched' but the pseudocode does not explicitly show how the generation, critique, and correction calls are batched; clarifying the batch structure would help implementation and compute accounting.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claims are self-contained and no fitted parameter is renamed as a prediction.

full rationale

The paper's central claim is an empirical comparison of a fixed, verifier-free pipeline (N=8 rollouts, D=4 refinement cycles, plurality vote) against standard baselines. The only derived quantity, Eq. (2), is a motivational bound on single-trace accuracy using latent accuracy λ and perturbation h adopted from Liu et al. (2024); these constants are not fitted to the reported benchmark numbers and do not enter the pipeline or the accuracy computation. No self-citations are load-bearing: all references to prior refinement or self-correction work are to external groups, and the method does not invoke a uniqueness theorem or an ansatz from the authors' own prior work. Hyperparameters N, D, and τ are fixed before the main evaluation; the AMC23 grid in Table 3 is an exploratory allocation study, and the authors explicitly note that accuracy is not monotonic in N or D, so the reported configuration is not the argmax of a fitted curve in any equation. The admitted lack of an equal-compute comparison (Section 5.3: the normalized gain 'does not constitute a direct equal-budget comparison') is a strength-of-evidence limitation, not a circular reduction: the accuracy numbers are not equal to the compute numbers by construction. The diagnostics in Figures 3–4 and Appendix A.8 also acknowledge negative per-rollout correction benefit, which is consistent with an honest empirical report rather than a claim forced by definition.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No fitted constants are used inside the method's decision rule; the main accuracy claim is empirical. Free parameters are hyperparameters N, D, and tau whose selection is not fully specified. The method introduces no new entities, only role-conditioned reuse of the same model.

free parameters (4)
  • Number of rollouts N = 8
    Main experiments use N=8; chosen as a moderate setting after the grid in Table 3, without a stated held-out validation protocol.
  • Refinement depth D = 4
    Main experiments use D=4; Section 5.3 says N=8, D=4 is a stable trade-off based on the AMC23 grid, which raises a post-hoc selection concern.
  • Sampling temperature tau = 0.7
    Set by hand for all sampling-based methods; standard but not justified beyond convention.
  • Clustering similarity threshold tau_sim = 0.85
    Used only for the redundancy diagnostics in Figure 1 and Appendix A.6; not central to the accuracy claims.
assumptions (5)
  • domain assumption Role-conditioned prompts turn the same base model into generator, critic, and corrector without additional training.
    Algorithm 1 and Section 3.3 assume the base model can follow these role prompts; if self-critique is unreliable, refinement adds noise, as acknowledged in the Limitations section.
  • domain assumption Majority voting over N refined rollouts suppresses isolated correct-to-wrong regressions.
    Section 5.2 and Appendix A.8 argue individual corrections are noisy but voting retains consistent improvements; this is asserted, not proven.
  • domain assumption The perturbation model in Eq. (2) describes per-trace hallucination and that width-only scaling cannot remove it.
    The inequality in Eq. (2) is algebraically correct, but the step from per-trace bias to a systematic accuracy floor for majority voting is asserted rather than derived.
  • domain assumption Qwen2.5-Math-RM-32B scores are a representative verifier baseline.
    RM@8 uses a single reward model; calibration issues are cited as motivation, but no sensitivity analysis over verifiers is provided.
  • standard math The standard transformer FLOP approximation (Kaplan and Hoffmann) adequately accounts for compute.
    Section 3.5 uses 2 times parameters times tokens for generation and prefill-only scoring; this ignores batching, memory, and prefill versus decode differences.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Refining Over Resampling: Test-Time Self-Correction for LLM Reasoning." pith.science (2026). https://pith.science/paper/UN4EWNYV

@misc{pith2026260805643,
  author       = {Pith},
  title        = {Pith review of: Refining Over Resampling: Test-Time Self-Correction for LLM Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UN4EWNYV}},
  note         = {Machine review of arXiv:2608.05643}
}
abstract

Test-time scaling improves LLM reasoning by using additional inference compute, but wider sampling alone can suffer from diminishing returns: new rollouts often repeat existing answer patterns instead of adding useful reasoning diversity. Verifier-based selection offers an alternative, but its performance depends on the calibration of an external reward model. We propose a verifier-free breadth--depth refinement framework that uses test-time compute to both explore and improve candidate solutions. The method samples multiple independent reasoning rollouts, refines each rollout through iterative self-critique and self-correction, and aggregates the refined answers by majority voting. Breadth preserves diverse initial attempts, while depth repairs local reasoning errors before aggregation. Across AIME24, AIME25, AMC, OlympiadBench, and MATH500, our method consistently improves over greedy decoding, majority voting, verifier-based best-of-$N$, beam search, and lookahead decoding across multiple open-weight models. For instance, with Qwen2.5-1.5B, accuracy increases from the strongest verifier-based baseline to $58.0\%$ on MATH500, and from $25.0\%$ to $32.5\%$ on AMC. These results show that test-time compute can be more effective when used to refine sampled trajectories rather than only to sample more candidates or rely on verifier-guided selection.

Figures

Figures reproduced from arXiv: 2608.05643 by the authors.

Figure 1
Figure 1. Sampling redundancy under increasing budget. As the sample budget grows, rollouts remain concentrated in a few semantic reasoning clusters, while the number of unique reasoning directions per sample decreases. We group initial rollouts into sentence￾embedding-based clusters by semantic similarity. Let {Cj}j≥1 be the clusters, with empirical mass pj (N) = N −1 PN i=1 1[r (0) i ∈ Cj ]. We track the number of realized … view at source ↗
Figure 2
Figure 2. Overview of the proposed breadth-depth refinement framework. Each of the N rollouts is refined through D depth layers, each consisting of three stages: (1) reasoning continuation (g (d) i ) by π G θ , (2) self-critique (c (d) i ) by π C θ , and (3) self-correction (r (d) i ) by π R θ . Terminal traces are aggregated by majority voting to produce the final answer aˆ. No verifier or additional training is required. pe… view at source ↗
Figure 3
Figure 3. Refinement dynamics across depths. Panels show recovery, regression, terminal rollout diversity, and answer agreement during refinement. 3.5 Compute Budget and TFLOP Accounting We estimate inference compute using the standard transformer scaling-law approximation (Kaplan et al., 2020; Hoffmann et al., 2022). For a method m, let Gm denote the set of autoregressive genera￾tion calls it performs, and let each call u ∈ … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Correction behavior across depths. Panels report answer changes, net correction benefit, and wrong-to￾correct versus correct-to-wrong flips during refinement. that critique provides useful intermediate feed￾back rather than merely increasing the token bud￾get. The effe…
Figure 5
Figure 5. Figure 5: Total inference compute in TFLOPs, shown on a log scale, for each method and benchmark across the [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Compute-normalized gain of structured refinement over each baseline. Bars report average accuracy￾point gain per additional 103 TFLOPs across the five benchmarks [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 12 canonical work pages

  1. [1]

    Reinforce LLM Reasoning through Multi-Agent Reflection

    Reinforce llm reasoning through multi-agent reflection , author=. arXiv preprint arXiv:2506.08379 , year=

    work page Pith review arXiv

  2. [2]

    arXiv preprint arXiv:2502.07803 , year=

    Reasoning-as-logic-units: Scaling test-time reasoning in large language models through logic unit alignment , author=. arXiv preprint arXiv:2502.07803 , year=

    arXiv

  3. [3]

    Think Thrice Before You Act: Progressive Thought Refinement in Large Language Models , url =

    Du, Chengyu and Han, Jinyi and Ying, Yizhou and Chen, Aili and He, Qianyu and Zhao, Haokun and Guo, Haoran and Xia, Sirui and Liang, Jiaqing and chen, zulong and Li, Liangyue and Xiao, Yanghua , booktitle =. Think Thrice Before You Act: Progressive Thought Refinement in Large Language Models , url =

    work page

  4. [4]

    International Conference on Learning Representations , volume=

    Progress or regress? self-improvement reversal in post-training , author=. International Conference on Learning Representations , volume=

    work page

  5. [5]

    International conference on learning representations , volume=

    Large language models cannot self-correct reasoning yet , author=. International conference on learning representations , volume=

  6. [6]

    2024 , eprint=

    Training Language Models to Self-Correct via Reinforcement Learning , author=. 2024 , eprint=

    2024

  7. [7]

    Advances in Neural Information Processing Systems , volume=

    Rest-mcts*: Llm self-training via process reward guided tree search , author=. Advances in Neural Information Processing Systems , volume=

  8. [8]

    2025 , eprint=

    TTRL: Test-Time Reinforcement Learning , author=. 2025 , eprint=

    2025

Show all 47 references
  1. [9]

    Advances in Neural Information Processing Systems , volume=

    Wider or deeper? scaling llm inference-time compute with adaptive branching tree search , author=. Advances in Neural Information Processing Systems , volume=

  2. [10]

    Peihao Wang and Ruisi Cai and Zhen Wang and Hongyuan Mei and qiang liu and Pan Li and Zhangyang Wang , booktitle=. \. 2026 , url=

    2026

  3. [11]

    arXiv preprint arXiv:2508.05004 , year=

    R-zero: Self-evolving reasoning llm from zero data , author=. arXiv preprint arXiv:2508.05004 , year=

    arXiv

  4. [12]

    arXiv preprint arXiv:2502.14565 , year=

    Revise: Learning to refine at test-time via intrinsic self-verification , author=. arXiv preprint arXiv:2502.14565 , year=

    arXiv

  5. [13]

    arXiv preprint arXiv:2507.12399 , year=

    ROC-n-reroll: How verifier imperfection affects test-time scaling , author=. arXiv preprint arXiv:2507.12399 , year=

  6. [14]

    2021 , eprint=

    Training Verifiers to Solve Math Word Problems , author=. 2021 , eprint=

    2021

  7. [15]

    The twelfth international conference on learning representations , year=

    Let's verify step by step , author=. The twelfth international conference on learning representations , year=

  8. [16]

    Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers) , pages=

    Making language models better reasoners with step-aware verifier , author=. Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1: long papers) , pages=

  9. [17]

    arXiv preprint arXiv:2501.07301 , year=

    The lessons of developing process reward models in mathematical reasoning , author=. arXiv preprint arXiv:2501.07301 , year=

    arXiv

  10. [18]

    Fixing the Broken Compass: Diagnosing and Improving Inference-Time Reward Modeling , author=

  11. [19]

    arXiv preprint arXiv:2203.11171 , year=

    Self-consistency improves chain of thought reasoning in language models , author=. arXiv preprint arXiv:2203.11171 , year=

    arXiv

  12. [20]

    arXiv preprint arXiv:2401.01879 , year=

    Theoretical guarantees on the best-of-n alignment policy , author=. arXiv preprint arXiv:2401.01879 , year=

    arXiv

  13. [21]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Trial and error: Exploration-based trajectory optimization of LLM agents , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  14. [22]

    arXiv preprint arXiv:2405.18208 , year=

    A human-like reasoning framework for multi-phases planning task with large language models , author=. arXiv preprint arXiv:2405.18208 , year=

    arXiv

  15. [23]

    arXiv preprint arXiv:2403.06963 , year=

    The pitfalls of next-token prediction , author=. arXiv preprint arXiv:2403.06963 , year=

    arXiv

  16. [24]

    arXiv preprint arXiv:2406.07394 , year=

    Accessing gpt-4 level mathematical olympiad solutions via monte carlo tree self-refine with llama-3 8b , author=. arXiv preprint arXiv:2406.07394 , year=

    arXiv

  17. [25]

    2023 , eprint=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. 2023 , eprint=

    2023

  18. [26]

    Advances in Neural Information Processing Systems , volume=

    Star: Bootstrapping reasoning with reasoning , author=. Advances in Neural Information Processing Systems , volume=

  19. [27]

    URL https://arxiv

    Reflexion: Language agents with verbal reinforcement learning, 2023 , author=. URL https://arxiv. org/abs/2303.11366 , volume=

    2023 arXiv

  20. [28]

    arXiv preprint arXiv:2211.00053 , year=

    Generating sequences by learning to self-correct , author=. arXiv preprint arXiv:2211.00053 , year=

    arXiv

  21. [29]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Math-shepherd: Verify and reinforce llms step-by-step without human annotations , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  22. [30]

    2024 , eprint=

    CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing , author=. 2024 , eprint=

    2024

  23. [31]

    International Conference on Learning Representations , volume=

    Teaching large language models to self-debug , author=. International Conference on Learning Representations , volume=

  24. [32]

    Advances in Neural Information Processing Systems , volume=

    Toward self-improvement of llms via imagination, searching, and criticizing , author=. Advances in Neural Information Processing Systems , volume=

  25. [33]

    International Conference on Learning Representations , volume=

    Enabling lanuguage models to implicitly learn self-improvement , author=. International Conference on Learning Representations , volume=

  26. [34]

    2024 , eprint=

    Large Language Models have Intrinsic Self-Correction Ability , author=. 2024 , eprint=

    2024

  27. [35]

    arXiv preprint arXiv:2409.12122 , year=

    Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement , author=. arXiv preprint arXiv:2409.12122 , year=

    arXiv

  28. [36]

    arXiv preprint arXiv:2407.10671 , year=

    Qwen2 Technical Report , author=. arXiv preprint arXiv:2407.10671 , year=

    arXiv

  29. [37]

    2024 , eprint=

    The Llama 3 Herd of Models , author=. 2024 , eprint=

    2024

  30. [38]

    2024 , eprint=

    OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems , author=. 2024 , eprint=

    2024

  31. [39]

    arXiv preprint arXiv:2103.03874 , year=

    Measuring mathematical problem solving with the math dataset , author=. arXiv preprint arXiv:2103.03874 , year=

    arXiv

  32. [40]

    2026 , eprint=

    Ministral 3 , author=. 2026 , eprint=

    2026

  33. [41]

    2024 , eprint=

    Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement , author=. 2024 , eprint=

    2024

  34. [42]

    Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

    Reimers, Nils and Gurevych, Iryna. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. 2019

    2019

  35. [43]

    arXiv preprint arXiv:2001.08361 , year=

    Scaling laws for neural language models , author=. arXiv preprint arXiv:2001.08361 , year=

    2001 arXiv

  36. [44]

    2022 , eprint=

    Training Compute-Optimal Large Language Models , author=. 2022 , eprint=

    2022

  37. [45]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Ranked voting based self-consistency of large language models , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  38. [46]

    2024 , eprint=

    MindStar: Enhancing Math Reasoning in Pre-trained LLMs at Inference Time , author=. 2024 , eprint=

    2024

  39. [47]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    Self-Refine: Iterative Refinement with Self-Feedback , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.