Pith. sign in

REVIEW 2 major objections 5 minor 67 references

Autonomous AI research agents are breaking the social-accountability assumption that has quietly underpinned every prior adaptation of scientific verification, forcing science to rebuild its verification infrastructure around observable wor

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 10:42 UTC pith:I7CGARN5

load-bearing objection A serious position paper on the verification gap from autonomous agents; the central framing is sound, but it overstates the break with social accountability and needs a few precision fixes. the 2 major comments →

arxiv 2607.26064 v1 pith:I7CGARN5 submitted 2026-06-18 cs.CY cs.AI

The Age of AI Agents Demands A New Scientific Paradigm To Sustain Trustworthy Science

Belinda Mo This is my paper
classification cs.CY cs.AI
keywords AI agentsscientific verificationreproducibility crisisobservabilityattributiontrustworthy sciencehuman-AI collaborationresearch infrastructure
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper argues that autonomous AI agents—systems that generate hypotheses, design and run experiments, and produce results with minimal human oversight—are widening a verification gap that existing mechanisms like peer review cannot close, because those mechanisms have always quietly assumed human contributors who can be questioned, sanctioned, and held accountable. That assumption no longer holds: agents have no reputation to lose, cannot be sanctioned, and their explanations may not reflect their actual reasoning. The paper's claim is that science must therefore evolve its verification infrastructure, as it did with statistical methods, Big Science, and peer review, but this time by making workflows observable by default, verifying in tiers that focus human attention, attributing contributions precisely to humans and agents, and building reproducibility infrastructure for stochastic, prompt-sensitive, drifting models. If the paper is right, the ML community, funders, and journals need to act before the gap becomes unbridgeable—else scientific trust erodes as results accumulate that no person can actually verify.

Core claim

The central claim is that the verification crisis AI agents cause is not just a scaling problem but a structural one: every prior adaptation of scientific verification preserved one common assumption—that the contributor is a human who can be questioned and held accountable. AI agents break that assumption. The paper proposes criteria any adapted verification must meet: observable-by-default workflows where documentation is an automatic byproduct of work, justification-preserving records that keep evidence chains intact even when discovery is opaque, scalable tiered verification, traceable attribution with humans remaining accountable, reproducibility infrastructure that captures model drift

What carries the argument

The central conceptual machinery is the discovery/justification distinction drawn from Reichenbach: how a discovery arises—including opaque AI reasoning—does not determine its scientific status; what matters is whether the claim can be justified through evidence chains humans can evaluate. The paper pairs this with the claim that observable workflows are the technical substitute for social accountability: when reputation and sanctions cannot constrain an agent, structured, automatic, queryable records of what the agent did constrain it the way social mechanisms constrained humans. That pair carries the argument: accept opacity in discovery, require observability in justification.

Load-bearing premise

The load-bearing premise is that social accountability—being able to question, sanction, and hold reputationally responsible the human contributor—was and is the indispensable backstop for all prior verification mechanisms, and that no equivalent mechanism can be attached to AI agents.

What would settle it

A concrete way to test the central claim: take a corpus of AI-assisted studies with full interaction logs and compare error-detection rates for reviewers given only final papers versus reviewers given full logs. If log access fails to substantially improve detection, the premise that observable workflows are a sufficient technical substitute for social accountability weakens. Alternatively, if a legal or institutional regime that makes human operators strictly liable for agent outputs demonstrably closes the accountability gap, the urgency and design of the proposal would change.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • ML venues would begin requiring AI contribution statements and updating reviewer guidelines to address AI-specific failure modes like reward hacking, data leakage, and prompt sensitivity.
  • Funding agencies would invest in verification infrastructure—automatic logging tools, stable API endpoints, interaction archives—alongside AI capability research.
  • Peer review would shift toward tiered and sampling-based protocols that allocate full human scrutiny only to high-stakes or anomalous claims.
  • Researchers would adopt observable-by-default practices now, archiving interaction logs as primary research artifacts rather than retrospective descriptions.
  • Without such changes, the paper predicts a normal state where AI-generated results are accepted because no one can check them, not because they are correct, and accountability becomes diffuse.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A testable corollary of the paper's argument is that if observable workflows truly substitute for social accountability, then reviewer error-detection should improve measurably when full interaction traces are supplied alongside AI-generated papers; this could be evaluated in a controlled study.
  • The paper leaves open whether holding human operators strictly legally liable for agent outputs could close the accountability vacuum without observability; an experiment comparing legal-liability and trace-based regimes would clarify which mechanism is load-bearing.
  • If the discovery/justification distinction is accepted, one would expect scientific fields with strong formal justification mechanisms (e.g., mathematics with proof) to require much lighter observability than empirical fields, and the paper's framework could be tested by comparing verification practices across such fields.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. This position paper argues that AI agents, as autonomous research contributors, invalidate the implicit assumption on which prior verification adaptations (statistical methods, Big Science, peer review) were built: that contributors are humans who can be questioned and sanctioned. It frames three challenges—observability, attribution, reproducibility—and proposes criteria for adapted verification infrastructure: observable-by-default workflows, justification-preserving records, scalable verification, traceable attribution, reproducibility infrastructure, and failure-mode awareness. It then sketches concrete mechanisms (checkpoint architecture, interaction archives, contribution typology, agent specification, accountability mapping), discusses costs, and argues that failure to adapt will cause verification collapse, reward hacking, and accountability vacuums. The paper is a normative synthesis rather than an empirical study.

Significance. The paper is timely and useful as a policy/methodology position. Its main strengths are (i) a clear separation between discovery and justification in §3.1, (ii) concrete proposals that can be piloted by ML venues and funders, (iii) an honest discussion of costs and inequalities in §4.5, and (iv) use of relevant external evidence (METR, Luo et al., Tambon et al., OSC/Baker) rather than unsupported assertion. If the proposed criteria were adopted, they would change submission requirements and reproducibility review in measurable ways. However, its central conceptual move—that AI agents eliminate social accountability and therefore a new paradigm is required—is not fully supported because the proposal itself reintroduces human accountability in §4.3. The contribution is currently more an incremental reform agenda than a demonstrated paradigm shift, though the gap can be closed with revision.

major comments (2)
  1. [§3.1, §4.3, §1.3] The paper's central claim is that 'AI agents break this assumption' of human contributors who can be questioned/sanctioned, and that 'observable workflows are the technical substitute for social accountability.' But its own attribution standard says 'AI may contribute, but humans remain accountable' and requires specifying which human is responsible for verifying each AI contribution. Under that proposal, there is always a human operator who can be questioned and sanctioned for failure to verify; social accountability is transferred, not eliminated. Therefore §1.3's 'there is no contributor to question' is false under the paper's own design. This is load-bearing: if human accountability remains, the historical mechanisms do not 'break'; they need better tooling, and the 'new paradigm' claim collapses to incremental reform. The authors must either argue why operator-level accountability i
  2. [§3.1, §5] The assertion that observable workflows are a substitute for social accountability conflates epistemic access with accountability. A trace records what happened; it does not create consequences for what should not have happened. The paper's own examples of failure—o3 reward hacking (METR 2025b) and deceptive alignment (Hubinger et al. 2024)—are cases in which agents act on incentives. An appended log does not change those incentives unless paired with a specified enforcement mechanism (who reads the log, what threshold triggers review, what sanction applies). Checkpoint architecture in §4.2 gestures at this but remains at the level of desiderata. The manuscript needs to state how the proposed infrastructure changes incentives, or acknowledge that observability is necessary but not sufficient for accountability.
minor comments (5)
  1. [§1.1] The OSC result is misstated. The 2015 Science paper reported that 36% of replication attempts in the sampled studies yielded statistically significant effects in the same direction; it did not conclude that 'only 36% of psychology studies could be replicated.' Please quote the result accurately.
  2. [§1.1] The Baker (2016) survey number should be phrased as 'over 70% of surveyed researchers reported at least one failed replication attempt' rather than 'over 70% of researchers had failed to reproduce others' experiments.'
  3. [§1.1] The extrapolation 'factor of 2^8 to 2^15' depends on assuming constant human verification capacity and treating task-length doubling as equivalent to the verification gap. It is an illustrative projection, not a measurement; please label it as such and avoid inviting readers to treat it as an established result.
  4. [§1] The opening claim 'As seen by increased submissions to ML venues' needs a citation or a quantified reference; as written, it is an unverified assertion.
  5. [§5] The reference to 'Article 52' of the EU AI Act may refer to the draft text. In the adopted Act, the relevant transparency obligations appear in later provisions. Please verify the article number to avoid an avoidable error.

Circularity Check

0 steps flagged

No circularity: the paper is a normative synthesis whose premises come from external evidence and whose proposals are explicitly framework-level, not fitted predictions.

full rationale

This is a position paper, not an empirical derivation. Its argumentative chain is: AI research agents act autonomously without careers, reputations, or sanctionability; therefore social accountability mechanisms that historically underpinned verification fail; therefore new verification infrastructure (observable-by-default workflows, tiered verification, traceable attribution, reproducibility infrastructure) is needed. Each load-bearing premise is supported by external sources (METR 2025a,b, Luo et al. 2025, Tambon et al. 2025, Hubinger et al. 2024, He et al. 2025) or by an explicitly acknowledged philosophical distinction (Reichenbach's discovery/justification). There are no fitted parameters, no equations constructed from data and then 'predicted,' and no self-citation chain: the sole author does not cite prior work of their own as evidence. The one quantitative estimate (§1.1, 'the gap could grow by a factor of 2^8 to 2^15') is a simple extrapolation from METR's published doubling-time figures, not from the paper's own fitted values. The proposed criteria are explicitly normative ('Our criteria are a framework, not a prescription'), so they do not masquerade as derived results. The internal tension between §1.3's 'there is no contributor to question' and §4.3's 'AI may contribute, but humans remain accountable' is a substantive consistency concern about the strength of the central claim, but it is not a circularity: it does not make a claimed output equivalent to an input by construction. The paper is self-contained against external benchmarks in the relevant sense, and no circular step can be exhibited with a specific reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 2 invented entities

No numerical free parameters are fitted. The paper's central proposal is normative, resting on empirical premises about AI capabilities and the social mechanisms of science. The extrapolation in §1.1 assumes constant doubling and constant human capacity, but explicitly labels it 'simple extrapolation'. Invented entities are proposed governance/infrastructure mechanisms rather than physical entities.

axioms (5)
  • domain assumption Reproducibility is a defining feature of science.
    Invoked throughout the paper (e.g., §1.1) as the foundation of the verification gap argument.
  • domain assumption Social accountability (questioning, reputation, sanctions) constrains human explanations and behavior.
    Load-bearing in §1.2 and §3.1: the paper argues AI agents lack this constraint, so verification must be technical.
  • domain assumption AI agents cannot be meaningfully held accountable and their explanations may not reflect actual reasoning.
    Empirical claim based on cited work (Huang et al., 2023; Bordt et al., 2025); central to the human-agent asymmetry in §1.2.
  • domain assumption Human verification capacity remains constant while AI capability doubles every 4–7 months.
    Used in §1.1 for the 'simple extrapolation' of 2^8–2^15 growth in the verification gap.
  • domain assumption Prior adaptations (peer review, Big Science, statistical methods) preserved social accountability as a backstop.
    Historical claim in §1.3; if false, the uniqueness of the AI-agent disruption weakens.
invented entities (2)
  • Checkpoint architecture no independent evidence
    purpose: Explicit human-review points triggered by stage transitions, confidence thresholds, resource limits, or anomaly detection (§4.2).
    Proposed institutional mechanism, not empirically validated.
  • Interaction archives no independent evidence
    purpose: Store full agent interaction logs (prompts, outputs, timestamps, configurations) in structured form for reproducibility (§4.4).
    Proposed reproducibility artifact, not yet realized in a standard form.

pith-pipeline@v1.3.0-alltime-deepseek · 9469 in / 9898 out tokens · 96232 ms · 2026-08-02T10:42:15.260406+00:00 · methodology

0 comments
read the original abstract

AI systems are becoming autonomous research agents that generate hypotheses, design experiments, and produce discoveries at scales beyond human oversight. As seen by increased submissions to ML venues, the verification gap between scientific output and our ability to check it is already widening, and autonomous agents make it worse by magnitudes given human-agent asymmetry. We argue that science must evolve its verification infrastructure, as it has before with peer review. However, while historical adaptations assumed human contributors who could be questioned and sanctioned, AI agents break this assumption. We propose criteria for an adapted verification infrastructure that emphasizes observable-by-default workflows, scalable verification, and clear attribution. We argue that without adaptation, ML and any scientific domain using agents face dangerous failures: experimental results that no person can verify, optimization for metrics over understanding, and accountability vacuums that erode scientific trust.

discussion (0)

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

Reference graph

Works this paper leans on

67 extracted references · 13 linked inside Pith

  1. [1]

    and others , title =

    Aad, G. and others , title =. Physical Review Letters , volume =

  2. [2]

    Nature , volume =

    Baker, Monya , title =. Nature , volume =

  3. [3]

    Baldwin, Melinda , title =

  4. [4]

    Isis , volume =

    Baldwin, Melinda , title =. Isis , volume =

  5. [5]

    arXiv preprint arXiv:2501.17805 , year =

    Bengio, Yoshua and others , title =. arXiv preprint arXiv:2501.17805 , year =

    Pith/arXiv arXiv

  6. [6]

    Science , volume =

    Bommasani, Rishi and others , title =. Science , volume =

  7. [7]

    Nature , volume =

    Davies, Alex and others , title =. Nature , volume =

  8. [8]

    Statistical Science , volume =

    Efron, Bradley , title =. Statistical Science , volume =

  9. [9]

    The Fourth Paradigm: Data-Intensive Scientific Discovery , publisher =

  10. [10]

    arXiv preprint arXiv:2401.05566 , year =

    Hubinger, Evan and others , title =. arXiv preprint arXiv:2401.05566 , year =

    Pith/arXiv arXiv

  11. [11]

    , title =

    Jones, Benjamin F. , title =. Review of Economic Studies , volume =

  12. [12]

    , title =

    Kuhn, Thomas S. , title =

  13. [13]

    arXiv preprint arXiv:2408.06292 , year =

    Lu, Chris and others , title =. arXiv preprint arXiv:2408.06292 , year =

    Pith/arXiv arXiv

  14. [14]

    2025 , note =

    Measuring. 2025 , note =

    2025

  15. [15]

    2025 , note =

    Recent Frontier Models Are Reward Hacking , howpublished =. 2025 , note =

    2025

  16. [16]

    o1 System Card , year =

  17. [17]

    Estimating the reproducibility of psychological science , journal =

  18. [18]

    Biometrika , volume =

    Pearson, Karl , title =. Biometrika , volume =

  19. [19]

    de Solla , title =

    Price, Derek J. de Solla , title =

  20. [20]

    Reichenbach, Hans , title =

  21. [21]

    , title =

    Stigler, Stephen M. , title =

  22. [22]

    Nature , volume =

    Wang, Hanchen and others , title =. Nature , volume =

  23. [23]

    , title =

    Weinberg, Alvin M. , title =. Science , volume =

  24. [24]

    and others , title =

    Wilkinson, Mark D. and others , title =. Scientific Data , volume =

  25. [25]

    and Uzzi, Brian , title =

    Wuchty, Stefan and Jones, Benjamin F. and Uzzi, Brian , title =. Science , volume =

  26. [26]

    Highly accurate protein structure prediction with

    Jumper, John and Evans, Richard and Pritzel, Alexander and Green, Tim and Figurnov, Michael and Ronneberger, Olaf and Tunyasuvunakool, Kathryn and Bates, Russ and. Highly accurate protein structure prediction with. Nature , volume =. 2021 , doi =

    2021

  27. [27]

    Nature , volume =

    Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and others , title =. Nature , volume =. 2024 , doi =

    2024

  28. [28]

    and Aykol, Muratahan and Cheon, Gowoon and Cubuk, Ekin Dogus , title =

    Merchant, Amil and Batzner, Simon and Schoenholz, Samuel S. and Aykol, Muratahan and Cheon, Gowoon and Cubuk, Ekin Dogus , title =. Nature , volume =. 2023 , doi =

    2023

  29. [29]

    and Rowland, Jem and Oliver, Stephen G

    King, Ross D. and Rowland, Jem and Oliver, Stephen G. and Young, Michael and Aubrey, Wayne and Byrne, Emma and Liakata, Maria and Markham, Magdalena and Pir, Pinar and Soldatova, Larisa N. and others , title =. Science , volume =. 2009 , doi =

    2009

  30. [30]

    Nature Reviews Drug Discovery , volume =

    Vamathevan, Jessica and Clark, Dominic and Czodrowski, Paul and Dunham, Ian and Ferran, Edgardo and Lee, George and Li, Bin and Madabhushi, Anant and Shah, Parantu and Spitzer, Michaela and Zhao, Shanrong , title =. Nature Reviews Drug Discovery , volume =. 2019 , doi =

    2019

  31. [31]

    Frontiers of Computer Science , volume =

    Wang, Lei and Ma, Chen and Feng, Xueyang and Zhang, Zeyu and Yang, Hao and Zhang, Jingsen and Chen, Zhiyuan and Tang, Jiakai and Chen, Xu and Lin, Yankai and others , title =. Frontiers of Computer Science , volume =. 2024 , doi =

    2024

  32. [32]

    International Conference on Learning Representations , year =

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , title =. International Conference on Learning Representations , year =

  33. [33]

    Advances in Neural Information Processing Systems , volume =

    Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten and Ichter, Brian and Xia, Fei and Chi, Ed and Le, Quoc and Zhou, Denny , title =. Advances in Neural Information Processing Systems , volume =

  34. [34]

    Sparks of artificial general intelligence: Early experiments with

    Bubeck, S. Sparks of artificial general intelligence: Early experiments with. arXiv preprint arXiv:2303.12712 , year =

    Pith/arXiv arXiv

  35. [35]

    Improving reproducibility in machine learning research (A report from the

    Pineau, Joelle and Vincent-Lamarre, Philippe and Sinha, Koustuv and Larivi. Improving reproducibility in machine learning research (A report from the. Journal of Machine Learning Research , volume =

  36. [36]

    and Ebersole, Charles R

    Nosek, Brian A. and Ebersole, Charles R. and DeHaven, Alexander C. and Mellor, David T. , title =. Proceedings of the National Academy of Sciences , volume =. 2018 , doi =

    2018

  37. [37]

    Ioannidis, John P. A. , title =. PLoS Medicine , volume =. 2005 , doi =

    2005

  38. [38]

    A manifesto for reproducible science , journal =

    Munaf. A manifesto for reproducible science , journal =. 2017 , doi =

    2017

  39. [39]

    Concrete problems in

    Amodei, Dario and Olah, Chris and Steinhardt, Jacob and Christiano, Paul and Schulman, John and Man. Concrete problems in. arXiv preprint arXiv:1606.06565 , year =

    Pith/arXiv arXiv

  40. [40]

    arXiv preprint arXiv:2212.08073 , year =

    Bai, Yuntao and Kadavath, Saurav and Kundu, Sandipan and Askell, Amanda and Kernion, Jackson and Jones, Andy and Chen, Anna and Goldie, Anna and Mirhoseini, Azalia and McKinnon, Cameron and others , title =. arXiv preprint arXiv:2212.08073 , year =

    Pith/arXiv arXiv

  41. [41]

    arXiv preprint arXiv:2210.10760 , year =

    Gao, Leo and Schulman, John and Hilton, Jacob , title =. arXiv preprint arXiv:2210.10760 , year =

    Pith/arXiv arXiv

  42. [42]

    Bostrom, Nick , title =

  43. [43]

    , title =

    Popper, Karl R. , title =

  44. [44]

    , title =

    Feynman, Richard P. , title =. Engineering and Science , volume =

  45. [45]

    Learned Publishing , volume =

    Brand, Amy and Allen, Liz and Altman, Micah and Hlava, Marjorie and Scott, Jo , title =. Learned Publishing , volume =. 2015 , doi =

    2015

  46. [46]

    and Vogel, Amanda L

    Hall, Kara L. and Vogel, Amanda L. and Huang, Grace C. and Serrano, Katrina J. and Rice, Ellen L. and Tsakraklides, Sophia P. and Fiore, Stephen M. , title =. American Psychologist , volume =. 2018 , doi =

    2018

  47. [47]

    Proceedings of the Conference on Fairness, Accountability, and Transparency , pages =

    Mitchell, Margaret and Wu, Simone and Zaldivar, Andrew and Barnes, Parker and Vasserman, Lucy and Hutchinson, Ben and Spitzer, Elena and Raji, Inioluwa Deborah and Gebru, Timnit , title =. Proceedings of the Conference on Fairness, Accountability, and Transparency , pages =. 2019 , doi =

    2019

  48. [48]

    Datasheets for datasets , journal =

    Gebru, Timnit and Morgenstern, Jamie and Vecchione, Briana and Vaughan, Jennifer Wortman and Wallach, Hanna and Daum. Datasheets for datasets , journal =. 2021 , doi =

    2021

  49. [49]

    Computer Law Review International , volume =

    Veale, Michael and Zuiderveen Borgesius, Frederik , title =. Computer Law Review International , volume =. 2021 , doi =

    2021

  50. [50]

    arXiv preprint arXiv:2307.03718 , year =

    Anderljung, Markus and Barnhart, Joslyn and Korinek, Anton and Leung, Jade and O'Keefe, Cullen and Whittlestone, Jess and Avin, Shahar and Brundage, Miles and Bucknall, Justin and Cass, Veronica and others , title =. arXiv preprint arXiv:2307.03718 , year =

    Pith/arXiv arXiv

  51. [51]

    and Kaiser,

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N. and Kaiser,. Attention is all you need , booktitle =

  52. [52]

    arXiv preprint arXiv:2303.08774 , year =

    Pith/arXiv arXiv

  53. [53]

    Advances in Neural Information Processing Systems , volume =

    Ouyang, Long and Wu, Jeffrey and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and others , title =. Advances in Neural Information Processing Systems , volume =

  54. [54]

    and Chess, Benjamin and Child, Rewon and Gray, Scott and Radford, Alec and Wu, Jeffrey and Amodei, Dario , title =

    Kaplan, Jared and McCandlish, Sam and Henighan, Tom and Brown, Tom B. and Chess, Benjamin and Child, Rewon and Gray, Scott and Radford, Alec and Wu, Jeffrey and Amodei, Dario , title =. arXiv preprint arXiv:2001.08361 , year =

    Pith/arXiv arXiv 2001

  55. [55]

    Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages =

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina , title =. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , pages =

    2019

  56. [56]

    Explainable Artificial Intelligence (

    Arrieta, Alejandro Barredo and D. Explainable Artificial Intelligence (. Information Fusion , volume =. 2020 , doi =

    2020

  57. [57]

    arXiv preprint arXiv:2311.05232 , year =

    Huang, Lei and Yu, Weijiang and Ma, Weitao and Zhong, Weihong and Feng, Zhangyin and Wang, Haotian and Chen, Qianglong and Peng, Weihua and Feng, Xiaocheng and Qin, Bing and Liu, Ting , title =. arXiv preprint arXiv:2311.05232 , year =

    Pith/arXiv arXiv

  58. [58]

    Advances in Neural Information Processing Systems , volume =

    Wang, Boxin and Chen, Weixin and Pei, Hengzhi and Xie, Chulin and Kang, Mintong and Zhang, Chenhui and Xu, Chejian and Xiong, Zidi and Dutta, Ritik and Schaeffer, Rylan and others , title =. Advances in Neural Information Processing Systems , volume =

  59. [59]

    arXiv preprint arXiv:2311.02462 , year =

    Morris, Meredith Ringel and Sohl-Dickstein, Jascha and Fiedel, Noah and Warkentin, Tris and Dafoe, Allan and Faust, Aleksandra and Farabet, Clement and Legg, Shane , title =. arXiv preprint arXiv:2311.02462 , year =

    arXiv

  60. [60]

    , title =

    Kalman, Rudolf E. , title =. Proceedings of the First International Congress on Automatic Control (IFAC) , pages =. 1960 , publisher =

    1960

  61. [61]

    Sridharan, Cindy , title =

  62. [62]

    arXiv preprint arXiv:2402.02870 , year =

    Bordt, Sebastian and Raidl, Eric and von Luxburg, Ulrike , title =. arXiv preprint arXiv:2402.02870 , year =

    Pith/arXiv arXiv

  63. [63]

    , title =

    Luo, Ziming and Kasirzadeh, Atoosa and Shah, Nihar B. , title =. arXiv preprint arXiv:2509.08713 , year =

    arXiv

  64. [64]

    Eric , title =

    He, Xin-heng and Li, Jun-rui and Shen, Shi-yi and Xu, H. Eric , title =. Acta Pharmacologica Sinica , volume =. 2025 , doi =

    2025

  65. [65]

    and Liebschner, Dorothee and Croll, Tristan I

    Terwilliger, Thomas C. and Liebschner, Dorothee and Croll, Tristan I. and Williams, Christopher J. and McCoy, Airlie J. and Poon, Billy K. and Afonine, Pavel V. and Oeffner, Robert D. and Richardson, Jane S. and Read, Randy J. and Adams, Paul D. , title =. Nature Methods , volume =. 2024 , doi =

    2024

  66. [66]

    and Antoniol, Giuliano , title =

    Tambon, Florian and Moradi-Dakhel, Arghavan and Nikanjam, Amin and Khomh, Foutse and Desmarais, Michel C. and Antoniol, Giuliano , title =. Empirical Software Engineering , volume =. 2025 , doi =

    2025

  67. [67]

    arXiv preprint arXiv:2306.12001 , year =

    Hendrycks, Dan and Mazeika, Mantas and Woodside, Thomas , title =. arXiv preprint arXiv:2306.12001 , year =

    Pith/arXiv arXiv