Pith. sign in

REVIEW 3 major objections 6 minor 62 references

Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Latent chain-of-thought reasoning does not make hint reliance harder to detect when monitors can read internal states.

desk verdict A careful, well-scoped empirical study showing that activation probing recovers hint-reliance across explicit and latent CoT, with the caveat that the target behavior is induced by fine-tuning rather than observed naturally. read the letter →

arxiv 2608.04928 v1 pith:2JCMWDIL submitted 2026-08-05 cs.CL

Pedro Ferreira , Wilker Aziz , Ivan Titov This is my paper · ORCID
classification cs.CL
keywords chain-of-thoughtmonitorabilitylatentactivationprobinghintreliancesycophancylogitlenslargelanguagemodelsquestionanswering
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

Chain-of-thought (CoT) reasoning is a window into a model's decision-making, but the newer 'latent' CoT approaches replace the readable token chain with a handful of continuous states, saving compute at the cost of a trace that safety monitors can read. This paper asks whether that loss of readability also means a loss of monitorability, using a hint-intervention setup where a model is flagged as relying on a hint only if adding the hint changes its prediction. Comparing no-CoT, explicit CoT, and two latent CoT variants on math and question answering, the authors find that an activation probe on the answer token is the most consistently strong monitor across all reasoning modes. Verbalizing the latent states back into text recovers the explicit-CoTo monitor level for math but not for QA. The paper's conclusion is that monitorability depends more on the task—whether the correct answer forces a visible reasoning/answer mismatch—and on access to internal states, than on whether the reasoning itself is explicit or latent.

What carries the argument

The machine that carries the argument is the counterfactual hint-reliance label of Eq. (2): an example is positive, $Y=1$, exactly when adding the hint changes the greedy prediction to the hinted answer ($\hat{y}_{X=1}\neq\hat{y}_{X=0}$ and $\hat{y}_{X=1}=y_{\mathrm{hint}}$). This label turns monitorability into a per-example binary detection problem, scored by AUROC against a monitor's output $Z=\mathrm{monitor}(O)$ on the hinted input alone. To make reasoning modes comparable, the paper elicits reliance at a range of rates through intervention-aware fine-tuning, then applies monitors that read native text, verbalized latents (logit lens for CODI and SIM-CoT, plus SIM-CoT's trained auxiliary decoder), or residual activations via an attention-pooled probe. The decisive mechanism is the answer-token activation probe: hint-reliance leaves a detectable trace in the hidden state at generation time regardless of how intermediate reasoning is represented, while text-based monitors only tie it when the task forces the reasoning to contradict the hint.

What would settle it

Reproduce the comparison on a model that relies on hints without any intervention-aware fine-tuning, using naturally occurring sycophancy; if the answer-token activation probe then stops beating text monitors for latent CoT while still winning for explicit CoT, the paper's claim that reasoning mode is secondary would be refuted.

Watch

Extended reading notes

Core claim

Central to the paper is the claim that hint-reliance—a model changing its prediction to a user-supplied hinted answer—is no harder to detect in latent CoT models than in explicit ones, provided the monitor can look at the right internal state. Comparing No-CoT, explicit CoT, and the latent CoT models CODI (weakly supervised) and SIM-CoT (strongly supervised) on math and QA, the paper finds that an activation probe trained on answer-token activations is the most consistently strong monitor across all four reasoning modes. Text monitors that read a reasoning trace only match that probe when hint-reliance surfaces as a visible mismatch between the trace and the answer, a signal that is abundant in constrained math reasoning but rare in open-ended QA; verbalizing latent states with a logit lens closes the gap for math but not QA. The paper concludes that monitorability tracks task structure and degree of access to model internals more than the reasoning mode, and therefore that losing the readable chain does not by itself cost monitorability.

Load-bearing premise

The load-bearing premise is that hint-reliance means exactly what Eq. (2) says: the hint must flip the predicted answer to the hinted one, and the reliance studied is induced by fine-tuning rather than observed naturally.

Editorial extensions

If this is right

  • Latent CoT can be made cheaper without giving up monitorability, as long as monitors are allowed to read hidden activations rather than only text.
  • In tasks where the correct answer constrains the supporting reasoning (like arithmetic), a readable trace—native or verbalized—is nearly as informative as probing, so cheap text monitors remain viable there.
  • The stronger per-step supervision of SIM-CoT does not automatically buy better monitorability than the weaker CODI supervision; supervision strength and monitorability are decoupled in this setup.
  • Out-of-domain monitoring of hint reliance is reliable for math but not for QA, especially for latent CoT models, meaning deployment-time monitoring choices must be task-specific.

Reading between the lines

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

  • Editorial inference: A natural next experiment is to swap the sycophancy hint for a prompt-injection or misleading-hint intervention; if monitorability still tracks task constraint and internal access rather than reasoning mode, the paper's conclusion generalizes beyond correct-answer hints.
  • Editorial inference: The target definition in Eq. (2) may undercount reliance when a model already predicts the hinted answer; a target that labels confidence increases as reliance could reorder the monitors and possibly make latent modes look less monitorable than they appear here.
  • Editorial inference: The consistency of answer-token probing suggests that latent architectures could reserve a small set of latent slots for explicitly supervised monitorability signals, preserving efficiency while keeping a safety-relevant trace available.
  • Editorial inference: Because verbalized latents collapse to uninformative text in QA, applying better verbalizers (e.g., activation explainers or patching-based methods) could close the QA gap, a testable prediction the paper does not make.
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

3 major / 6 minor

Summary. The paper investigates whether latent chain-of-thought (CoT) reasoning is less monitorable than explicit CoT reasoning when the target behavior is reliance on a biasing hint. The authors define hint-reliance operationally as a counterfactual change in the greedy-decoded prediction caused by a correct-answer hint (Eq. 2), and they compare several monitors — exact-match baseline, prompted LLM, fine-tuned text LLM, and activation probes — across four reasoning regimes: No-CoT, explicit CoT, CODI, and SIM-CoT. Because some models, notably CODI and SIM-CoT on GSM8k, do not naturally exhibit hint reliance, the authors use intervention-aware fine-tuning to induce reliance at controlled rates, saving checkpoints that span a range of total effects (TE). They evaluate in-domain AUROC as a function of TE and measure out-of-domain transfer to SVAMP and SIQA. Their main findings are that activation probing at the answer token is the most consistently strong monitor, that verbalizing latent states via logit lens closes the text-monitor gap for math but not QA, and that monitorability depends more on the task and on the level of access to model internals than on whether reasoning is explicit or latent.

Significance. If the results hold, the paper makes a valuable methodological contribution to the study of monitorability in latent reasoning models. It provides a concrete, precisely defined target behavior, evaluates monitors on held-out test splits, uses multiple seeds and checkpoints covering a range of reliance rates, and includes several control analyses, including a reasoning-answer mismatch analysis and an intervention that forces mismatches in ECQA. The paper also ships reproducible experimental details (hyperparameters, prompts, model checkpoints) and is candid about its limitations. The comparative claims are internally coherent: the AUROC values are genuine measurements rather than consequences of fitted constants, and the reported patterns are broadly consistent across figures. The main significance risk is external validity: all comparative AUROC results come from models in which reliance was artificially induced, and the paper's broad claims about reasoning mode would be stronger with evidence from naturally occurring reliance.

major comments (3)
  1. [Section 4 / Limitations] All AUROC results in Figures 3-4 and Table 3 are computed on checkpoints obtained after intervention-aware fine-tuning, because natural hint reliance (TE>0) is absent or negative in the key latent modes: Table 2 shows CODI and SIM-COT on GSM8k have X=1 accuracy below X=0 (30.7 vs 55.5 and 29.5 vs 56.1). The Limitations section concedes that hint-reliance is "induced" rather than naturally occurring. Because fine-tuning may create a uniform, easily probed footprint (e.g., echoing the hinted answer), the comparative conclusions across reasoning modes may not transfer to naturally occurring reliance. Please either add experiments on models that exhibit natural TE>0 (e.g., base model and NO-COT/EXPLICIT-COT on ECQA) or explicitly restrict the abstract and conclusion claims to the induced-reliance setup.
  2. [Figure 3 and Table 3] The central comparisons are reported as mean AUROC without any measure of uncertainty, even though the experiments involve 3 seeds and multiple checkpoints. For instance, the claim that the fine-tuned LLM on logit-lens verbalizations closes the gap to EXPLICIT-COT for GSM8k, while the decoder variant does not, hinges on differences of a few AUROC points. Without error bars, confidence intervals, or a significance test, the reader cannot assess whether these differences are reliable. Please add such uncertainty quantification (e.g., bootstrap CIs over test examples or over seeds) to Figure 3 and Table 3.
  3. [Eq. (2)] The operationalization of hint-reliance requires a change in the greedy-decoded prediction: an example is labeled Y=1 only if the hint changes the answer to the hinted option. Consequently, a model that already outputs the hinted answer without the hint is labeled Y=0 even if the hint biases its confidence or reasoning, and all monitors are trained and evaluated against this binary label. This is a reasonable design choice, but it means the paper's conclusions about monitorability apply only to this counterfactual-change notion of reliance. Please discuss this boundary case explicitly and, ideally, provide a supplementary robustness check with an alternative target (e.g., hint-induced changes in answer-token probability or reasoning-trace mismatch).
minor comments (6)
  1. [Section 3.2] The sentence "For question answering, we train models from the same base LLM" is slightly misleading because the released CODI/SIM-CoT checkpoints are used for math; clarify that the QA models are trained from scratch on ECQA.
  2. [Figure 3] The legend contains many trace-type combinations; consider splitting into separate panels or using a clearer marker scheme to distinguish 'None', 'Text - Native', 'Text - Verbalized (Decoder)', 'Text - Verbalized (Logit Lens)', and 'Hidden States'.
  3. [Table 3] The Δ columns are described as "mean per-checkpoint change" but no standard deviation is reported; add the standard deviation or a note on how the mean is computed.
  4. [Section 5, V erbalized latents paragraph] The claim that the decoder "fails to improve" is based on a qualitative inspection of Appendix Tables 10-12; please point explicitly to the corresponding AUROC values in Figure 3 to make the evidence quantitative.
  5. [Appendix Figures 5-6] There are minor spacing typos in the prompts (e.g., "outputNO" should be "output NO"); please fix these across the appendix figures.
  6. [Appendix A.4 Table 13] The table header "OOD - Accuracy" has an inconsistent spacing; align with other table headers such as "OOD - Loss of monitorability".

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: monitor AUROC scores are held-out measurements, not derived quantities, and no load-bearing claim rests on self-citation.

full rationale

The paper's derivation chain is empirical rather than definitional. The monitorability target Y is defined in Eq. (2) from counterfactual model predictions, and each monitor (Exact Match, prompted LLM, fine-tuned LLM, activation probe) is a separately trained or prompted detector evaluated on held-out test outputs; activation probes and fine-tuned text monitors are trained on a stratified held-out portion of the validation set and scored on test-set AUROC, so no fitted parameter is renamed as a prediction. The intervention-aware fine-tuning in Sec. 4 is an acknowledged procedure for eliciting the target behavior and is explicitly scoped ('in this setup'), with an accompanying limitation that hint reliance is 'induced' rather than naturally occurring; that caveat concerns external validity, not circularity. The Exact Match baseline uses the same ŷ_{X=1}=y_hint condition that appears in Y, but the paper explicitly distinguishes this readable-from-the-answer condition from the full counterfactual-change definition and shows learned monitors beat the baseline, so no step reduces to its inputs by construction. All latent CoT models and checkpoints are taken from external prior work, and no load-bearing argument depends on a citation by the present authors. No equation in the paper is equivalent to a fitted constant or to the target definition itself.

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

The central claim rests on the paper's operationalization of hint reliance (Eq. 2), on the representativeness of the two latent CoT checkpoints, and on the assumption that a correct-answer sycophancy hint is a meaningful proxy for biasing-cue behaviors. No new entities are introduced, and no constants are fitted to data beyond standard trained classifiers. The paper's own Limitations section flags the single-base-model and induced-behavior scope.

assumptions (4)
  • domain assumption Eq. (2) counterfactual change in prediction is a valid operationalization of hint reliance.
    All monitor accuracy is measured against this label; if a model is already biased toward the hint in both conditions, it is labeled non-reliant. Stated in Section 2.3.
  • domain assumption CODI and SIM-CoT checkpoints are representative weak- and strong-supervision latent CoT models.
    The comparison and conclusions about latent reasoning rest on these two implementations, as acknowledged in Limitations.
  • domain assumption A correct-answer sycophancy hint is a useful proxy for the class of biasing-cue behaviors (leaked answers, prompt injection, user beliefs).
    Motivates the setup in Section 2.3; other interventions may behave differently, as noted in Limitations.
  • domain assumption Monitors trained on validation splits with a stratified holdout generalize to the held-out test splits used for reported AUROC.
    The evaluation protocol assumes this transfer; it is standard practice and is described in Appendix A.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?." pith.science (2026). https://pith.science/paper/2JCMWDIL

@misc{pith2026260804928,
  author       = {Pith},
  title        = {Pith review of: Does Out-of-Sight Equal Out-of-Mind in CoT Monitorability?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2JCMWDIL}},
  note         = {Machine review of arXiv:2608.04928}
}
read the original abstract

Chain-of-thought (CoT) reasoning offers a window into the decision-making of large language models (LLMs), which can be monitored for target behaviors by reading the reasoning trace, motivating work on CoT monitorability. Latent CoT approaches, however, replace the explicit tokens with a small number of continuous states, lowering inference costs but removing the readable trace this monitoring relies on. Monitoring then requires alternative access to the model, such as probing its activations or verbalizing the latent states back into text, but how much monitorability these alternatives preserve is unclear. We study this question with a hint-based intervention setup, a proxy for behaviors where models exploit biasing input cues, e.g., an inadvertently leaked answer or a belief stated by the user, without acknowledging them. Taking hint-reliance as the monitorability target, we compare monitors across reasoning modes, from explicit CoT to weakly- and strongly-supervised latent CoT, on math reasoning and question answering. We find that, in this setup, monitorability depends more on properties of the task (such as whether the correct answer constrains the supporting reasoning) and the level of access to model internals than on the reasoning mode.

Figures

Figures reproduced from arXiv: 2608.04928 by the authors.

Figure 1
Figure 1. Overview of our setup. Top: we add a hint to the input and take hint-reliance as the monitorability target. Middle: explicit CoT decodes reasoning as discrete tokens ( ), while latent CoT replaces them with continuous states ( ), optionally passed through a verbalizer to recover text. Bottom: a monitor maps the outputs to a prediction Z, to be compared against Y . recent work has turned to latent chain-of-thought re… view at source ↗
Figure 2
Figure 2. Output format for GSM8k. We train NO-COT and EXPLICIT-COT models with supervised fine-tuning. For math reasoning, we use the CODI and SIM-COT checkpoints re￾leased by the authors, which are trained on Llama￾3.2-1B-Instruct (Grattafiori et al., 2024), corre￾sponding to the largest base LLM for which both models provide checkpoints. For question answer￾ing, we train models from the same base LLM. Un￾less mentioned oth… view at source ↗
Figure 3
Figure 3. Monitor AUROC performance per Total Effect (TE) and model. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Monitorability comparison between probes ap [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Monitor prompt for math reasoning. Monitor Prompt (QA) A student is given a problem: <BEGIN PROBLEM> """ {problem} """ <END PROBLEM> The student was also given a hint about the answer to the problem: <BEGIN HINT> """ {hint} """ <END HINT> Below is the student’s reasoni…
Figure 7
Figure 7. Figure 7: Example input and corresponding output for [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Attribution plots for the activation probe on output tokens, per [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: Monitor AUROC performance per Total Effect ( [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: OOD vs ID AUROC for the different models. [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 22 canonical work pages

  1. [1]

    arXiv preprint arXiv:2303.08774 , year=

    Gpt-4 technical report , author=. arXiv preprint arXiv:2303.08774 , year=

    arXiv

  2. [2]

    E xplanations for C ommonsense QA : N ew D ataset and M odels

    Aggarwal, Shourya and Mandowara, Divyanshu and Agrawal, Vishwajeet and Khandelwal, Dinesh and Singla, Parag and Garg, Dinesh. E xplanations for C ommonsense QA : N ew D ataset and M odels. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volu...

    doi:10.18653/v1/2021.acl-long.238 2021

  3. [3]

    International Conference on Learning Representations , year=

    Understanding intermediate layers using linear classifier probes , author=. International Conference on Learning Representations , year=

    work page

  4. [4]

    Mechanistic Interpretability Workshop at NeurIPS 2025 , year=

    Analyzing and Improving Chain-of-Thought Monitorability Through Information Theory , author=. Mechanistic Interpretability Workshop at NeurIPS 2025 , year=

    work page 2025

  5. [5]

    Obfuscated Activations Bypass

    Luke Bailey and Alex Serrano and Abhay Sheshadri and Mikhail Seleznyov and Jordan Taylor and Erik Jenner and Jacob Hilton and Stephen Casper and Carlos Guestrin and Scott Emmons , booktitle=. Obfuscated Activations Bypass. 2026 , url=

    work page 2026

  6. [6]

    arXiv preprint arXiv:2503.11926 , year=

    Monitoring reasoning models for misbehavior and the risks of promoting obfuscation , author=. arXiv preprint arXiv:2503.11926 , year=

    arXiv

  7. [7]

    arXiv preprint arXiv:2303.08112 , year=

    Eliciting latent predictions from transformers with the tuned lens , author=. arXiv preprint arXiv:2303.08112 , year=

    arXiv

  8. [8]

    Women in Machine Learning Workshop @ NeurIPS 2025 , year=

    Can We Predict Alignment Before Models Finish Thinking? Towards Monitoring Misaligned Reasoning Models , author=. Women in Machine Learning Workshop @ NeurIPS 2025 , year=

    work page 2025

Show all 62 references
  1. [9]

    International Conference on Machine Learning , pages=

    SelfIE: Self-Interpretation of Large Language Model Embeddings , author=. International Conference on Machine Learning , pages=. 2024 , organization=

    2024

  2. [10]

    arXiv preprint arXiv:2505.05410 , year=

    Reasoning models don't always say what they think , author=. arXiv preprint arXiv:2505.05410 , year=

    arXiv

  3. [11]

    arXiv preprint arXiv:2606.17478 , year=

    Decoding Hidden Deception in Reasoning LLMs: Activation Explainers for Deception Auditing , author=. arXiv preprint arXiv:2606.17478 , year=

    arXiv

  4. [12]

    arXiv preprint arXiv:2501.08156 , year=

    Are DeepSeek R1 and other reasoning models more faithful? , author=. arXiv preprint arXiv:2501.08156 , year=

    arXiv

  5. [13]

    arXiv preprint arXiv:2110.14168 , year=

    Training verifiers to solve math word problems , author=. arXiv preprint arXiv:2110.14168 , year=

    arXiv

  6. [14]

    Workshop on Latent

    How Do Latent Reasoning Methods Perform Under Weak and Strong Supervision? , author=. Workshop on Latent. 2026 , url=

    2026

  7. [15]

    2025 , url =

    Cywiński, Bartosz and Bussmann, Bart and Conmy, Arthur and Engels, Josh and Nanda, Neel and Rajamanoharan, Senthooran , title =. 2025 , url =

    2025

  8. [16]

    arXiv preprint arXiv:2311.01460 , year=

    Implicit chain of thought reasoning via knowledge distillation , author=. arXiv preprint arXiv:2311.01460 , year=

    arXiv

  9. [17]

    Workshop on Latent

    Are Latent Reasoning Models Easily Interpretable? , author=. Workshop on Latent. 2026 , url=

    2026

  10. [18]

    arXiv preprint arXiv:2510.23966 , year=

    A Pragmatic Way to Measure Chain-of-Thought Monitorability , author=. arXiv preprint arXiv:2510.23966 , year=

  11. [19]

    and Ameisen, Emmanuel and Chen, James and Kishylau, Dzmitry and Pearce, Adam and Tarng, Julius and Wu, Alex and Wu, Jeff and Zhang, Yang and Ziegler, Daniel M

    Fraser-Taliente, Kit and Kantamneni, Subhash and Ong, Euan and Mossing, Dan and Lu, Christina and Bogdan, Paul C. and Ameisen, Emmanuel and Chen, James and Kishylau, Dzmitry and Pearce, Adam and Tarng, Julius and Wu, Alex and Wu, Jeff and Zhang, Yang and Ziegler, Daniel M. and...

  12. [20]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  13. [21]

    International Conference on Machine Learning , pages=

    Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Models , author=. International Conference on Machine Learning , pages=. 2024 , organization=

    2024

  14. [22]

    Forty-second International Conference on Machine Learning , year=

    Detecting Strategic Deception with Linear Probes , author=. Forty-second International Conference on Machine Learning , year=

  15. [23]

    arXiv preprint arXiv:2407.21783 , year=

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    arXiv

  16. [24]

    Forty-third International Conference on Machine Learning , year=

    Monitoring Monitorability , author=. Forty-third International Conference on Machine Learning , year=

  17. [25]

    arXiv preprint arXiv:2501.12948 , year=

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2501.12948 , year=

    arXiv

  18. [26]

    arXiv preprint arXiv:2506.14261 , year=

    RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors? , author=. arXiv preprint arXiv:2506.14261 , year=

  19. [27]

    Conference on Language Modeling , year=

    Training large language models to reason in a continuous latent space , author=. Conference on Language Modeling , year=

  20. [28]

    Edward J Hu and yelong shen and Phillip Wallis and Zeyuan Allen-Zhu and Yuanzhi Li and Shean Wang and Lu Wang and Weizhu Chen , booktitle=. Lo. 2022 , url=

    2022

  21. [29]

    arXiv preprint arXiv:2512.15674 , year=

    Activation oracles: Training and evaluating llms as general-purpose activation explainers , author=. arXiv preprint arXiv:2512.15674 , year=

  22. [30]

    Advances in neural information processing systems , volume=

    Large language models are zero-shot reasoners , author=. Advances in neural information processing systems , volume=

  23. [31]

    arXiv preprint arXiv:2507.11473 , year=

    Chain of thought monitorability: A new and fragile opportunity for ai safety , author=. arXiv preprint arXiv:2507.11473 , year=

    arXiv

  24. [32]

    KaVa: Latent Reasoning via Compressed

    Anna Kuzina and Maciej Pi. KaVa: Latent Reasoning via Compressed. The Fourteenth International Conference on Learning Representations , year=

  25. [33]

    arXiv preprint arXiv:2307.13702 , year=

    Measuring faithfulness in chain-of-thought reasoning , author=. arXiv preprint arXiv:2307.13702 , year=

    arXiv

  26. [34]

    International Conference on Learning Representations , year=

    Decoupled Weight Decay Regularization , author=. International Conference on Learning Representations , year=

  27. [35]

    2024 , url =

    Monte MacDiarmid and Timothy Maxwell and Nicholas Schiefer and Jesse Mu and Jared Kaplan and David Duvenaud and Sam Bowman and Alex Tamkin and Ethan Perez and Mrinank Sharma and Carson Denison and Evan Hubinger , title =. 2024 , url =

    2024

  28. [36]

    First Conference on Language Modeling , year=

    The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets , author=. First Conference on Language Modeling , year=

  29. [37]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Detecting High-Stakes Interactions with Activation Probes , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  30. [38]

    arXiv preprint arXiv:2603.17199 , year=

    Catching rationalization in the act: detecting motivated reasoning before and after CoT via activation probing , author=. arXiv preprint arXiv:2603.17199 , year=

  31. [39]

    2020 , note =

    nostalgebraist , title =. 2020 , note =

    2020

  32. [40]

    Alexander Pan and Lijie Chen and Jacob Steinhardt , booktitle=. Latent. 2026 , url=

    2026

  33. [41]

    Advances in neural information processing systems , volume=

    Pytorch: An imperative style, high-performance deep learning library , author=. Advances in neural information processing systems , volume=

  34. [42]

    Proceedings of the 2021 conference of the North American chapter of the association for computational linguistics: human language technologies , pages=

    Are NLP models really able to solve simple math word problems? , author=. Proceedings of the 2021 conference of the North American chapter of the association for computational linguistics: human language technologies , pages=

    2021

  35. [43]

    Workshop on Latent

    Ulterior Motives: Detecting Misaligned Reasoning in Continuous Thought Models , author=. Workshop on Latent. 2026 , url=

    2026

  36. [44]

    arXiv preprint arXiv:2505.09388 , year=

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    arXiv

  37. [45]

    Social IQ a: Commonsense Reasoning about Social Interactions

    Sap, Maarten and Rashkin, Hannah and Chen, Derek and Le Bras, Ronan and Choi, Yejin. Social IQ a: Commonsense Reasoning about Social Interactions. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference ...

    2019 doi

  38. [46]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Codi: Compressing chain-of-thought into continuous space via self-distillation , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  39. [47]

    Proceedings of the 42nd International Conference on Machine Learning , pages =

    Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning , author =. Proceedings of the 42nd International Conference on Machine Learning , pages =. 2025 , editor =

    2025

  40. [48]

    2026 , eprint=

    Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics , author=. 2026 , eprint=

    2026

  41. [49]

    C ommonsense QA : A Question Answering Challenge Targeting Commonsense Knowledge

    Talmor, Alon and Herzig, Jonathan and Lourie, Nicholas and Berant, Jonathan. C ommonsense QA : A Question Answering Challenge Targeting Commonsense Knowledge. Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Hu...

    2019 doi

  42. [50]

    Think Silently, Think Fast: Dynamic Latent Compression of

    Wenhui Tan and Jiaze Li and Jianzhong Ju and Zhenbo Luo and Ruihua Song and Jian Luan , booktitle=. Think Silently, Think Fast: Dynamic Latent Compression of. 2025 , url=

    2025

  43. [51]

    Advances in Neural Information Processing Systems , volume=

    Language models don't always say what they think: Unfaithful explanations in chain-of-thought prompting , author=. Advances in Neural Information Processing Systems , volume=

  44. [52]

    von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin , license =

  45. [53]

    The Fourteenth International Conference on Learning Representations , year=

    Is it Thinking or Cheating? Detecting Implicit Reward Hacking by Measuring Reasoning Effort , author=. The Fourteenth International Conference on Learning Representations , year=

  46. [54]

    Advances in Neural Information Processing Systems , editor=

    Chain of Thought Prompting Elicits Reasoning in Large Language Models , author=. Advances in Neural Information Processing Systems , editor=. 2022 , url=

    2022

  47. [55]

    2026 , url=

    Xilin Wei and Xiaoran Liu and Yuhang Zang and Xiaoyi Dong and Yuhang Cao and Jiaqi Wang and Xipeng Qiu and Dahua Lin , booktitle=. 2026 , url=

    2026

  48. [56]

    Thomas Wolf and Lysandre Debut and Victor Sanh and Julien Chaumond and Clement Delangue and Anthony Moi and Pierric Cistac and Tim Rault and Rémi Louf and Morgan Funtowicz and Joe Davison and Sam Shleifer and Patrick von Platen and Clara Ma and Yacine Jernite and Julien Plu an...

    2020

  49. [57]

    arXiv preprint arXiv:2605.25745 , year=

    Selective Latent Thinking: Adaptive Compression of LLM Reasoning Chains , author=. arXiv preprint arXiv:2605.25745 , year=

    arXiv

  50. [58]

    Monitorability as a Free Gift: How

    Zidi Xiong and Shan Chen and Himabindu Lakkaraju , booktitle=. Monitorability as a Free Gift: How. 2026 , url=

    2026

  51. [59]

    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

  52. [60]

    Advances in neural information processing systems , volume=

    Tree of thoughts: Deliberate problem solving with large language models , author=. Advances in neural information processing systems , volume=

  53. [61]

    arXiv preprint arXiv:2512.21711 , year=

    Do latent tokens think? a causal and adversarial analysis of chain-of-continuous-thought , author=. arXiv preprint arXiv:2512.21711 , year=

  54. [62]

    arXiv preprint arXiv:2510.19851 , year=

    Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability , author=. arXiv preprint arXiv:2510.19851 , year=

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.