Governing AI Beyond the Pretraining Frontier

Nicholas A. Caputo

REVIEW 2 major objections 5 minor 2 cited by

Governing AI Beyond the Pretraining Frontier

T0 review · 2 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Frontier AI governance, built on pretraining compute thresholds, will become misaligned if the pretraining paradigm ends; the paper proposes transparency, new bottlenecks, and regulatory capacity as the path forward.

desk verdict A serious, clearly written policy essay whose actionable conclusion is conditional on an empirical bet that pretraining scaling is ending; the abstract overstates the evidence, but the scenario analysis and concrete governance proposals are genuinely useful. read the letter →

arxiv 2502.15719 v1 pith:KZCB5PGM submitted 2025-01-27 cs.CY cs.AI

Nicholas A. Caputo This is my paper · ORCID

classification cs.CYcs.AI

keywords frontierAIregulationpretrainingparadigmcomputethresholdsscalinglawssafetytransparencyinference-timedatawall

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

Frontier AI laws now being enacted in the EU, US, UK, and China treat pretraining scale as the choke point that makes the technology governable: compute thresholds (like the EU AI Act's $10^{25}$ FLOPs presumption) and chip export controls assume that more pretraining compute is the main route to more capability. The paper argues that this "pretraining paradigm" is hitting a wall because good training data is finite and recent large runs disappointed their developers, while capabilities keep advancing through inference-time reasoning and other routes. If the paradigm ends, the paper claims, the regulatory field will become more diffuse: more companies can reach the "pretraining frontier," new bottlenecks replace compute, and current triggers stop tracking risk. The paper introduces the "pretraining frontier" as the capability ceiling of pretraining alone and proposes a path forward based on transparency, new natural bottlenecks (data, inference compute, algorithmic innovation), and stronger regulatory capacity. The stakes are immediate because several major legal frameworks come into force this year and may be misaligned with the technology as it actually develops.

What carries the argument

The object that carries the argument is the "pretraining frontier": the capabilities threshold made possible by scaling up pretraining alone, given finite high-quality data and compute. It does the work of defining what changes after the pretraining paradigm: it marks the point at which scaling no longer buys capability, opening the field to more actors and forcing regulation to target new bottlenecks. The other load-bearing mechanism is the compute-threshold trigger (for example, the EU AI Act's $10^{25}$ FLOPs presumption and US Executive Order 14110's $10^{26}$ FLOPs threshold), which the paper analyzes as the legal expression of the pretraining assumption.

What would settle it

A single observation would settle much of the debate: if a next-generation model trained with substantially more compute (at or above the EU's FLOP threshold) delivers a clear capability jump over GPT-4-class models and continues to improve according to scaling laws, then the pretraining frontier is not binding and compute-based thresholds remain effective. The paper itself cites sources predicting this outcome, so the distinction is observable in upcoming model releases.

Watch

Extended reading notes

Core claim

The central claim is that current frontier AI governance is built on the pretraining paradigm—the regularity that scaling up pretraining data and compute produces predictable capability gains—and that this paradigm is ending, so the regulations built on it are becoming misaligned with the technology they are meant to govern. The load-bearing regulatory tools are compute thresholds: the EU AI Act's $10^{25}$ FLOPs presumption, US Executive Order 14110's $10^{26}$ FLOPs trigger, and US export controls on advanced microchips, all of which assume that large pretraining runs are the key input to frontier capability and the natural bottleneck to monitor, control, or exclude. The paper introduces the "pretraining frontier," the capabilities ceiling reachable by scaling pretraining alone, and argues that if it binds, the field deconcentrates as chip scarcity eases, more players reach the frontier, and new sources of progress—inference-time compute, synthetic data, and algorithmic innovation—take over. It then argues that regulators should stop betting on pretraining compute as the universal handle and instead invest in transparency, find new natural bottlenecks (data, inference compute, information), and build regulatory capacity, while protecting rights. If the argument is right, the legal order for frontier AI now being built is aimed at the wrong layer of the technology.

Load-bearing premise

The argument stands or falls on the premise that the pretraining paradigm is actually ending—that scaling pretraining with more data and compute no longer yields significant capability gains—rather than merely slowing temporarily.

Editorial extensions

If this is right

If the pretraining frontier binds, compute thresholds such as the EU's FLOP-based presumption and US chip export controls stop tracking capability, because compute no longer predicts gains.
More companies can reach the frontier as chip scarcity eases and Moore's law raises effective compute supply, making the frontier field more diffuse and harder to monitor.
Incumbents may preserve their lead by shifting to inference-time compute scaling and synthetic data, creating new bottlenecks that regulators can target.
Regulation should move toward transparency, monitoring inputs like data and inference compute, and grouping models by risk class rather than by training scale.
If capability gains continue through non-pretraining paths, regulators should count all compute used in development and deployment, not just pretraining compute.

Reading between the lines

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

A testable corollary of the paper's argument is that the distribution of frontier-capable models should widen over the next few years, with open-weight models reaching GPT-4-class performance even without massive new pretraining runs.
If the pretraining frontier is real, export controls on advanced chips lose strategic value while retaining diplomatic costs; the paper gestures at this but does not develop the geopolitical consequence.
The paper's logic implies that regulators should start building capability-evaluation triggers now, before a crisis makes them necessary, rather than waiting to see which bottleneck stabilizes.
Inference-time compute governance would depend on cloud providers monitoring customer usage, which raises privacy and civil-liberties tradeoffs that the paper acknowledges but does not resolve.

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.

Desk Editor's Note

read the letter

Colleague,

The one thing to know: this is a policy essay, not a technical paper, and the genuinely new piece is the 'pretraining frontier' construct—the idea that if pretraining scaling tops out, the regulatory choke points built on compute thresholds (EU AI Act Article 51, US export controls, the draft UK bill) lose their target. The paper also maps out a concrete menu of alternatives: inference-compute reporting by cloud providers, data stewardship, algorithmic-transparency regimes, and capacity-building. That synthesis goes beyond prior work (Sastry et al., Narayanan et al.) that flagged the general worry about compute thresholds becoming outdated.

The paper does well on structure and honesty. It acknowledges the empirical disagreement (refs 17 and 67 argue scaling-style progress continues via reasoning training infrastructure), and Section 4 explicitly lays out three futures—AI winter, technical breakthrough keeping pretraining dominant, and paradigm shift. In the first two, the paper concedes existing compute-based governance 'would likely be preserved.' That is the right kind of conditional honesty.

The soft spot is real and load-bearing. The abstract and the policy recommendations are written as though the paradigm shift is already happening, while the body only supports it as a scenario. The evidence for the end of pretraining is a handful of press reports that companies were disappointed with recent runs, plus the finite supply of human text. The paper itself notes the disagreement. No likelihood estimate or decision-theoretic framing is provided, so the claim that regulators must reorient now is not entailed by the analysis. A reader who assigns high probability to continued scaling—through reasoning models, synthetic data, or better training infrastructure—has little reason to adopt the proposed shift away from compute-centric triggers. This is not a flaw in internal logic; it is an empirical premise that the paper treats as more settled than its own footnotes allow. Framing the paper explicitly as scenario analysis would fix most of this.

Minor point: the section on the EU AI Act is strong on the Article 51 triggers but slightly overstates how much the 10^25 threshold will 'implicitly' drive coverage; the evaluations trigger and Commission decisions are genuine alternatives, and the paper acknowledges this but then downplays it.

Bottom line: for AI governance readers and policymakers, this is a valuable scenario-planning piece and deserves a serious referee; it should be revised to align the abstract with the conditional body. I'd bring it to a reading group and would cite it as an example of forward-looking regulatory analysis.

Recommendation: send to peer review, with the expectation of a major revision focused on reframing the empirical claim.

Referee Report

2 major / 5 minor

Summary. The paper is a policy analysis of frontier AI regulation in light of a possible end to the "pretraining paradigm"—the assumption that scaling up pretraining compute and data is the primary driver of AI capability gains. It argues that current regulatory instruments (the EU AI Act's 10^25 FLOPs threshold, US export controls on chips, the forthcoming UK approach, and Chinese draft AI law) are premised on pretraining scale as a legible, trackable bottleneck. If that paradigm ends, the paper contends, these regulations become misaligned with a more diffuse and capability-diverse frontier. The paper introduces the concept of the "pretraining frontier," analyzes industry-structure implications such as deconcentration, and proposes regulatory alternatives: increased transparency, data-oriented triggers, wider or inference-focused compute governance, selective information regulation, and capacity-building. The central argument is conditional: much of the text says "if the pretraining paradigm ends," and Section 4 explicitly lists alternative futures in which existing frameworks would remain effective.

Significance. The paper addresses a timely and important policy question with concrete, actionable proposals, and it is unusually honest about the uncertainty surrounding its central premise. It engages with the actual mechanisms of existing laws (compute thresholds, export controls, institutional capacity) and offers a useful taxonomy of possible regulatory responses. Its discussion of evaluating risks, grouping models by capabilities, and using data or inference-compute as alternative bottlenecks is thoughtful and goes beyond generic calls for 'more AI safety research.' The manuscript is not an empirical study; it does not claim to prove that pretraining scaling is ending, and much of the analysis is self-consciously conditional. That being said, the abstract states the paradigm shift as an unfolding fact, while the body concedes that the evidence is contested and that other futures are possible. If reframed to align its claims with its own conditional structure, the paper would be a valuable contribution to the frontier AI governance literature.

major comments (2)

[Abstract and Section 4] The abstract asserts an actual paradigm shift ('This paradigm shift presents fundamental challenges... threatening to undermine this new legal order as it emerges'), but Section 4 (paragraph 2) states that if progress stalls into an AI winter or if technical breakthroughs overcome the data wall, 'the effectiveness of existing regulatory frameworks would likely be preserved' because compute-based governance continues to work. The actionable conclusion that regulators should reorient now is therefore not entailed by the paper's own conditional analysis; it depends on an unstated probability judgment about which future is more likely. The manuscript should either consistently frame its thesis as conditional ('if the pretraining paradigm ends, then...') throughout the abstract and recommendations, or provide an explicit decision-theoretic justification for adapting regulations now despite the uncertainty. Without this, the abstract overstates what the analysis establishes.
[Section 2] The evidentiary basis for the load-bearing premise that the pretraining paradigm is ending is thin and the paper itself acknowledges it is contested. The supporting cites are press reports of disappointment at OpenAI, Google, and Anthropic (refs 52, 65, 78), an estimate of the finite supply of human text (ref 85), and a prediction by Ilya Sutskever (ref 82); the paper explicitly notes that refs 17 and 67 disagree, with SemiAnalysis arguing that scaling-law-style progress continues through reasoning and post-training infrastructure. If the manuscript aims to persuade the reader that the paradigm shift is actually occurring, this evidence is insufficient and should be strengthened with quantitative trends (e.g., scaling-law fits, benchmark improvements over compute, model release trajectories). If it is merely a conditional analysis, the premise should be presented as a stipulated scenario rather than as the basis for the abstract's factual-sounding claim.

minor comments (5)

[Title page] The manuscript is printed with 'Unpublished working draft. Not for distribution.' at the top of every page; for a journal submission this header should be removed and replaced with a conventional title page.
[Section 2, final paragraph] The sentence 'And regulations premised on the idea that scaling with remain the driver of capabilities will have to change.' contains a typo: 'scaling with remain' should read 'scaling will remain'.
[References] References are formatted inconsistently: [31] lacks the article title, [36] is given as a White House fact sheet but is cited for export controls that are also described in [33] and [35], and several URLs are broken across line breaks (e.g., [28], [66]). The list should be cleaned up before publication.
[Section 3.3] The analysis of the 'Forthcoming UK Frontier AI Bill' is necessarily provisional because the bill had not been released at the time of writing. The paper should state its 'as of' date more prominently in that section, since the analysis may quickly become dated if the bill is published with different triggers.
[Section 5.3] The proposal that cloud providers report inference-compute expenditures above a threshold is interesting but underspecified regarding legal authority, technical feasibility, and privacy implications; a footnote or caveat acknowledging these open questions would strengthen the presentation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: this is a conditional policy analysis resting on external legal texts and technical reports, with only one non-load-bearing self-citation and an explicit admission that alternative futures preserve existing frameworks.

full rationale

The paper does not derive a quantity, fit a parameter, or run a model; it critiques existing regulations (EU AI Act Article 51's 10^25 FLOPs trigger, EO 14110's compute thresholds, and US export controls) against a stipulated concept, the 'pretraining frontier.' The central empirical premise—that pretraining scaling is ending—is supported by external reporting (refs 52, 65, 78) and the paper explicitly acknowledges disagreement (refs 17, 67). Section 4 states that if progress stalls into an AI winter or if technical breakthroughs overcome the data wall, existing regulatory frameworks 'would likely be preserved' because compute governance continues to work. This is an honest conditional, not a conclusion smuggled in through definition: the recommendations are framed as responses to one possible future, not as logical consequences of the paper's own axioms. The only self-citation, ref 15 in Section 3, supports the uncontroversial observation that frontier regulators 'rely on similar tools and on cooperation with each other'; it is illustrative, not load-bearing. No uniqueness theorem is imported from the authors' prior work, no ansatz is smuggled in via citation, and no fitted input is relabeled as a prediction. The paper's own limitation language makes clear that the force of its argument depends on an empirical bet about the end of pretraining scale, which is a correctness or evidence concern, not circularity. Under the stated rules, this yields a score of 1: a minor self-citation exists but the central analysis is self-contained against external sources and does not reduce to its own inputs.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The paper rests on empirical and normative premises. The key empirical premise, the end of pretraining scaling, is disputed. The 'pretraining frontier' is a new conceptual entity but has no independent evidence. No free parameters are fitted because there is no quantitative model.

assumptions (4)

domain assumption Scaling laws let regulators forecast model capabilities from compute and data inputs.
Used to justify that compute thresholds give predictability (Section 2, citing refs 7, 34, 40).
domain assumption The pretraining paradigm is ending or will end soon.
Central empirical premise; supported by press reports and data estimates, but contested by refs 17 and 67 (Sections 2, 4).
domain assumption Frontier AI poses novel CBRN, cyber, persuasion, and loss-of-control risks that justify regulation.
Adopted from cited safety literature [5, 28, 41]; used as the motivation for the entire regulatory analysis.
ad hoc to paper Regulatory burdens should be minimized and fundamental rights protected.
Normative premise stated in the abstract and conclusion, shaping the evaluation of policy options.

invented entities (1)

pretraining frontier
purpose: A capabilities ceiling that scaling pretraining alone can reach, used to argue the regulatory field will become diffuse and competition will increase.
Introduced in Section 4.1 with no quantitative definition or falsifiable prediction; it is a conceptual construct, not a measurable empirical claim.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Governing AI Beyond the Pretraining Frontier." pith.science (2026). https://pith.science/paper/KZCB5PGM

@misc{pith2026250215719,
  author       = {Pith},
  title        = {Pith review of: Governing AI Beyond the Pretraining Frontier},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KZCB5PGM}},
  note         = {Machine review of arXiv:2502.15719}
}

read the original abstract

This year, jurisdictions worldwide, including the United States, the European Union, the United Kingdom, and China, are set to enact or revise laws governing frontier AI. Their efforts largely rely on the assumption that increasing model scale through pretraining is the path to more advanced AI capabilities. Yet growing evidence suggests that this "pretraining paradigm" may be hitting a wall and major AI companies are turning to alternative approaches, like inference-time "reasoning," to boost capabilities instead. This paradigm shift presents fundamental challenges for the frontier AI governance frameworks that target pretraining scale as a key bottleneck useful for monitoring, control, and exclusion, threatening to undermine this new legal order as it emerges. This essay seeks to identify these challenges and point to new paths forward for regulation. First, we examine the existing frontier AI regulatory regime and analyze some key traits and vulnerabilities. Second, we introduce the concept of the "pretraining frontier," the capabilities threshold made possible by scaling up pretraining alone, and demonstrate how it could make the regulatory field more diffuse and complex and lead to new forms of competition. Third, we lay out a regulatory approach that focuses on increasing transparency and leveraging new natural technical bottlenecks to effectively oversee changing frontier AI development while minimizing regulatory burdens and protecting fundamental rights. Our analysis provides concrete mechanisms for governing frontier AI systems across diverse technical paradigms, offering policymakers tools for addressing both current and future regulatory challenges in frontier AI.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

How to Catch a GPU: A Taxonomy of Verification and Enforcement Mechanisms for International AI Agreements
cs.CY 2026-06 conditional novelty 6.0 of 10

Verification of international AI agreements will fail first at detecting hidden compute facilities, around the 10,000-H100-equivalent scale, before other enforcement mechanisms break.
Meek Models Shall Inherit the Earth
cs.AI 2025-07 conditional novelty 5.0 of 10

Under fixed-distribution neural scaling laws, the capability gap between state-of-the-art and low-compute AI models shrinks over time toward zero.

Reference graph

Works this paper leans on

88 extracted references · 60 canonical work pages · cited by 2 Pith papers

[1]

European Union AI Act. 2024. Article 51: Classiﬁcation o f General-Purpose AI Models as General-Purpose AI Models wi th Systemic Risk. https://artiﬁcialintelligenceact.eu/article/51/

2024
[2]

European Union AI Act. 2024. Chapters III and V. https://artiﬁcialintelligenceact.eu/article/51/

2024
[3]

Meta AI. 2024. Introducing Meta Llama 3: The most capable openly available LLM to date. https://ai.meta.com/blog/meta-llama-3/

2024
[4]

UK AI Safety Institute. 2024. Inspect AI: Framework for Large Language Model Evaluations . https://github.com/UKGovernmentBEIS/inspect_ai

2024
[5]

Markus Anderljung, Joslyn Barnhart, Anton Korinek, Jad e Leung, Cullen O’Keefe, Jess Whittlestone, Shahar Avin, Mi les Brundage, Justin Bullock, Duncan Cass-Beggs, Ben Chang, Tantum Collins, Tim Fist, Gil lian Hadﬁeld, Alan Hayes, Lewis Ho, Sara Hooker, Eric Horvit z, Noam Kolt, Jonas Schuett, Yonadav Shavit, Divya Siddarth, Robert Trager, an d Kevin Wolf...

arXiv 2023
[6]

Gregory Arcuri and Sujai Shivakumar. 2022. Moore’s Law a nd Its Practical Implications.https://www.csis.org/analysis/moores-law-and-its-pr actical-implications. Center for Security and International Studies

2022
[7]

Yasaman Bahri, Ethan Dyer, Jared Kaplan, Jaehoon Lee, an d Utkarsh Sharma. 2024. Explaining neural scaling laws. Proceedings of the National Academy of Sciences 121, 27 (2024), e2311878121. doi:10.1073/pnas.2311878121 arXiv:https://www.pnas.org/doi/pdf/10.1073/pnas.2311878121

doi:10.1073/pnas.2311878121 2024
[8]

Arsheeya Bajwa and Krystal Hu. 2024. Amazon doubles down on AI startup Anthropic with another $4 bln. https://www.reuters.com/technology/artiﬁcial-intelligence/anthropic-receives-4-billion-investment-amaz on-makes-aws-oﬃcial-cloud-provider-2024-11-22 . Reuters

2024

Show all 88 references

[9]

Dean W. Ball. 2024. Here’s What I Think We Should Do. https://www.hyperdimensional.co/p/heres-what-i-thin k-we-should-do . Hyperdimen- sional

2024
[10]

Ball and Daniel Kokotajlo

Dean W. Ball and Daniel Kokotajlo. 2024. 4 Ways to Advanc e Transparency in Frontier AI Development. https://time.com/collection/time100-voices/7086285/ai-transparency-measures/ . Time

2024
[11]

United Nations AI Advisory Body. 2024. Governing AI for Humanity. https://www.un.org/sites/un2.un.org/ﬁles/governing_ai_for_humanity_ﬁnal_report_en.pdf

2024
[12]

Daniel James Bogiatzis-Gibbons. 2024. Beyond Individ ual Accountability: (Re-)Asserting Democratic Control of AI. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). Association for Computing Machinery, ...

2024
[13]

Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbia h, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pr anav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeﬀ...

2020 arXiv
[14]

Miles Brundage. 2025. https://x.com/Miles_Brundage/status/1876051275693044197. X (formerly Twitter)

2025
[15]

Caputo, Haydn Belﬁeld, Jakob Mökander, Mat teo Pistillo, Huw Roberts, Sophie Williams, and Robert Trager

Nicholas A. Caputo, Haydn Belﬁeld, Jakob Mökander, Mat teo Pistillo, Huw Roberts, Sophie Williams, and Robert Trager. 2024. International Governance through Domestic L aw in the Forthcoming UK Frontier AI Bill. https://www.oxfordmartin.ox.ac.uk/publications/international-gover...

2024
[16]

Francois Chollet. 2024. OpenAI o3 Breakthrough High Sc ore on ARC-AGI-Pub. https://arcprize.org/blog/oai-o3-pub-breakthrough . ARC Prize

2024
[17]

Jack Clark. 2024. Import AI 395: AI and energy demand; di stributed training via DeMo; and Phi-4. https://jack-clark.net/2024/12/23/import-ai-395-ai- and-energy-demand-distributed-training-via-demo-and -phi-4/. Import AI

2024
[18]

DeepSeek-AI. 2024. DeepSeek-V3. https://github.com/deepseek-ai/DeepSeek-V3

2024
[19]

DeepSeek-AI. 2025. DeepSeek-R1. https://github.com/deepseek-ai/DeepSeek-R1/blob/mai n/DeepSeek_R1.pdf

2025
[20]

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv: 1810.04805 [cs.CL] https://arxiv.org/abs/1810.04805

2019 arXiv
[21]

Nathalie Diberardino, Clair Baleshta, and Luke Stark. 2024. Algorithmic Harms and Algorithmic Wrongs. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). Association for Computing Machinery, New York, NY...

2024
[22]

European Commission Directorate-General for Communi cation. 2024. AI Act enters into force.https://commission.europa.eu/news/ai-act-enters-for ce-2024-08-01_en

2024
[23]

Michelle Donelan. 2023. Introducing the AI Safety Inst itute. https://assets.publishing.service.gov.uk/media/65438d159e05fd0014be7bd9/introducing-ai-safety-institut e-web-accessible.pdf . UK Department for Science, Innovation & Technology

2023
[24]

Clare Duﬀy. 2025. Trump announces a $500 billion AI infr astructure investment in the US.https://edition.cnn.com/2025/01/21/tech/openai-oracle-softbank-trump-ai-investment/index.html . CNN

2025
[25]

Janet Egan and Eric Rosenbach. 2023. Biosecurity in the Age of AI: What’s the Risk?https://www.belfercenter.org/publication/biosecurity-age-ai-whats-risk . Harvard Kennedy School Belfer Center for Science and Intern ational Aﬀairs. 15 Unpublished working draft.Not for distribu...

2023
[26]

UK Department for Science Innovation & Technology. 202 3. Capabilities and risks from frontier AI. https://assets.publishing.service.gov.uk/media/65395abae6c968000daa9b25/frontier-ai-capabilities-risks -report.pdf
[27]

UK Department for Science Innovation & Technology. 202 4. Frontier AI Safety Commitments, AI Seoul Summit 2024. https://www.gov.uk/government/publications/frontier-ai-safety-commitments-ai-seoul-summit-2024/frontie r-ai-safety-commitments-ai-seoul-summit-2024

2024
[28]

Frontier Model Forum. 2024. Issue Brief: Components of Frontier AI Safety Frameworks.https://www.frontiermodelforum.org/updates/issue-brief-components-of-frontier-ai-safety-

2024
[29]

Anissa Gardizy and Amir Efrati. 2024. Microsoft and Ope nAI Plot $100 Billion Stargate AI Supercomputer. https://www.theinformation.com/articles/microsoft-and-openai-plot-100-billion-stargate-ai-supercompute r. The Information

2024
[30]

Aaron Grattaﬁori, Abhimanyu Dubey, Abhinav Jauhri, Ab hinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letm an, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal , Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravank umar, Artem Ko- renev...

2024 arXiv
[31]

Anna Gross and Stephanie Stacy. 2024. UK will legislate against AI risks in next year, pledges Kyle. https://www.ft.com/content/79fedc1c-579d-4b23-8404- e4cb9e7bbae3. Financial Times

2024
[32]

Barath Harithas and Andreas Schumacher. 2024. Where th e Chips Fall: U.S. Export Controls Under the Biden Administr ation from 2022 to 2024. https://www.csis.org/analysis/where-chips-fall-us-e xport-controls-under-biden-administration-2022-2024 . Center for Strategic & Interna...

2024
[33]

Mackenzie Hawkins and Jenny Leonard. 2025. Biden to Fur ther Limit Nvidia AI Chip Exports in Final Push. https://www.bloomberg.com/news/articles/2025-01-08/ biden-to-further-limit-nvidia-amd-ai-chip-exports-i n-ﬁnal-push. Bloomberg

2025
[34]

Rae, Oriol Vi nyals, and Laurent Sifre

Jordan Hoﬀmann, Sebastian Borgeaud, Arthur Mensch, El ena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hen- dricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Nol and, Katie Millican, George van den Driessche, Bogdan Damoc , Aurelia Guy, Simon ...

2022 arXiv
[35]

Horowitz

Michael C. Horowitz. 2025. What to Know About the New U.S . AI Diﬀusion Policy and Export Controls. https://www.cfr.org/blog/what-know-about-new-us-ai- diﬀusion-policy-and-export-controls . Council on Foreign Relations

2025
[36]

The White House. 2025. FACT SHEET: Ensuring U.S. Securi ty and Economic Strength in the Age of Artiﬁcial Intelligenc e. https://www.whitehouse.gov/brieﬁng-room/statements-releases/2025/01/13/fact-sheet-ensuring-u-s-securit y-and-economic-strength-in-the-age-of-artiﬁcial-int elligence/

2025
[37]

US AI Safety Institute. 2024. The United States Artiﬁci al Intelligence Safety Institute: Vision, Mission, and Str ategic Goals. https://www.nist.gov/system/ﬁles/documents/2024/05/21/AISI-vision-21May2024.pdf . National Institute of Standards and Technology

2024
[38]

Sami Jaghouar, Jack Min Ong, Manveer Basra, Fares Obeid , Jannik Straube, Michael Keiblinger, Elie Bakouch, Lucas Atkins, Maziyar Panahi, Charles Goddard, Max Ryabinin, and Johannes Hagemann. 2024. INTELLECT-1 Technical Repor7. https://github.com/PrimeIntellect-ai/prime/blob/m...

2024
[39]

Nicola Jones. 2024. How should we test AI for human-leve l intelligence? OpenAI’s o3 electriﬁes quest. Nature (2024). doi:10.1038/d41586-024-00110-6

2024 doi
[40]

Brow n, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Je ﬀrey Wu, and Dario Amodei

Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brow n, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Je ﬀrey Wu, and Dario Amodei. 2020. Scaling Laws for Neural Language Models. arXi v:2001.08361 [cs.LG] https://arxiv.org/abs/2001.08361

2020 arXiv
[41]

Leonie Koessler, Jonas Schuett, and Markus Anderljung . 2024. Risk thresholds for frontier AI. arXiv: 2406.14713 [cs.CY] https://arxiv.org/abs/2406.14713

2024 arXiv
[42]

Scott Kohler and Ian Klaus. 2024. A Heated California De bate Oﬀers Lessons for AI Safety Governance. https://carnegieendowment.org/posts/2024/10/california-sb1047-ai-safety-bill-veto-lessons?lang=en . Carnegie Endowment for International Peace

2024
[43]

Michael Kremer. 1998. Patent Buyouts: A Mechanism for E ncouraging Innovation*. The Quarterly Journal of Economics 113, 4 (11 1998), 1137–1167. doi:10.1162/003355398555865 arXiv:https://academic.oup.com/qje/article-pdf/113/4/1137/5406836/113-4-1137.pdf

1998 doi
[44]

Nathan Lambert. 2024. OpenAI’s o3: The grand ﬁnale of AI in 2024. https://www.interconnects.ai/p/openais-o3-the-2024- ﬁnale-of-ai . Intercon- nects

2024
[45]

Nathan Lambert. 2025. DeepSeek R1’s recipe to replicat e o1 and the future of reasoning LMs. https://www.interconnects.ai/p/deepseek-r1-recipe-f or-o1. Interconnects

2025
[46]

Greg Leppert. 2024. How Knowledge Institutions Can Bui ld a Promethean Moment. https://institutionaldatainitiative.org/hello-world.html. In- stitutional Data Initiative

2024
[47]

Jiatong Li, Renjun Hu, Kunzhe Huang, Yan Zhuang, Qi Liu, Mengxiao Zhu, Xing Shi, and Wei Lin. 2024. PertEval: Unveili ng Real Knowledge Capacity of LLMs with Knowledge-Invariant Perturbations. arXiv:2405.19740 [cs.CL] https://arxiv.org/abs/2405.19740 17 Unpublished working dra...

2024 arXiv
[48]

Zhang Linghan, Yang Jianjun, Cheng Ying, Zhao Jingwu, H an Xuzhi, Zheng Zhifeng, and Xu Xiaoben. 2023. Artiﬁcial Int elligence Law of the Peo- ple’s Republic of China (Draft for Suggestions from Scholars). https://cset.georgetown.edu/wp-content/uploads/t0592_china_ai_law_draft...

2023
[49]

Shuo Liu, Kaining Ying, Hao Zhang, Yue Yang, Yuqi Lin, Ti anle Zhang, Chuanhao Li, Yu Qiao, Ping Luo, Wenqi Shao, and Ka ipeng Zhang
[50]

Thomas Maxwell and Hugh Langley. 2023. Google pulls bac k on how much AI research it publishes as ChatGPT eats its lunch: ’It’s time to compete’. https://www.businessinsider.com/google-publishing-less-conﬁdential-ai-research-to-compete-with-openai- 2023-4. Business Insider

2023
[51]

Sandra G. Mayson. 2019. Bias In, Bias Out. Yale Law Journal 128, 8 (2019)

2019
[52]

Rachel Metz, Shirin Ghaﬀary, Dina Bass, and Julia Love. 2024. OpenAI, Google and Anthropic Are Struggling to Build M ore Advanced AI. https://www.bloomberg.com/news/articles/2024-11-13/ openai-google-and-anthropic-are-struggling-to-build -more-advanced-ai . Bloomberg

2024
[53]

Piotr Mirowski, Juliette Love, Kory Mathewson, and Sha kir Mohamed. 2024. A Robot Walks into a Bar: Can Language Mode ls Serve as Creativity SupportTools for Comedy? An Evaluation of LLMs’ Humour Alig nment with Comedians. In Proceedings of the 2024 ACM Conference on Fair- nes...

2024
[54]

Howard Morland. 2005. Born Secret. Cardozo Law Review 26, 4 (2005)

2005
[55]

Arvind Narayanan, Benedikt Stroebl, and Sayash Kapoor . 2024. Reasoning Model Evaluations. https://benediktstroebl.github.io/reasoning-model-e vals/

2024
[56]

Arvind Narayanan, Benedikt Ströbl, and Sayash Kapoor. 2024. Is AI progress slowing down?https://www.aisnakeoil.com/p/is-ai-progress-slowing -down. AI Snake Oil

2024
[57]

Kwan Yee Ng, Jason Zhou, Ben Murphy, Rogier Creemers, an d Hunter Dorwart. 2023. Translation: Artiﬁcial Intelligence Law, Model Law v. 1.0 (Ex- pert Suggestion Draft) – Aug. 2023.https://digichina.stanford.edu/work/translation-artiﬁcial-intelligence-law-model-law-v-1-0-expert-...

2023
[58]

Cyberspace Administration of China. 2021. Provisions on the Management of Algorithmic Recommendations in Intern et Information Services. https://www.chinalawtranslate.com/en/algorithms/. China Law Translate

2021
[59]

Cyberspace Administration of China. 2022. Provisions on the Administration of Deep Synthesis Internet Informati on Services. https://www.chinalawtranslate.com/en/deep-synthesis/. China Law Translate

2022
[60]

Cyberspace Administration of China. 2023. Interim Mea sures for the Management of Generative Artiﬁcial Intellige nce Services. https://www.chinalawtranslate.com/en/generative-ai- interim/. China Law Translate

2023
[61]

US Bureau of Industry and Security. 2025. Biden-Harris Administration Announces Regulatory Framework for the Res ponsible Diﬀusion of Ad- vanced Artiﬁcial Intelligence Technology.https://www.bis.gov/press-release/biden-harris-admi nistration-announces-regulatory-framework-resp...

2025
[62]

Nuria Oliver, Alexander Peukert, Matthias Samwald, Yo shua Bengio, Marietje Schaake, Rishi Bommasani, Céline Cas tets-Renard, Marta Ziosi, Alexander Zacherl, Daniel Privitera, Nitarshan Rajkumar, Anka Reuel, and Markus Anderljung. 2024. Second Draft Gener al-Purpose AI Code of...

2024
[63]

OpenAI. 2024. OpenAI o1 and new tools for developers. https://openai.com/index/o1-and-new-tools-for-devel opers/

2024
[64]

Stephanie Palazzolo. 2025. OpenAI Preps ’Operator’ Re lease for this Week.https://www.theinformation.com/brieﬁngs/openai-preps-operator-release-for-this-week . The Information

2025
[65]

Stephanie Palazzolo, Erin Woo, and Amir Efrati. 2024. O penAI Shifts Strategy as Rate of ‘GPT’ AI Improvements Slows . https://www.theinformation.com/articles/openai-shifts-strategy-as-rate-of-gpt-ai-improvements-slows . The Information

2024
[66]

European Parliament. 2024. EU AI Act: ﬁrst regulation o n artiﬁcial intelligence. https://www.europarl.europa.eu/topics/en/article/20230601STO93804/eu-ai-act-ﬁrst-regulation-

2024
[67]

Failures

Dylan Patel, Daniel Nishball, and AJ Kourabi. 2024. Sca ling Laws – O1 Pro Architecture, Reasoning Training Infrast ructure, Orion and Claude 3.5 Opus “Failures”. https://semianalysis.com/2024/12/11/scaling-laws-o1 -pro-architecture-reasoning-training-infrastructure -orion-and...

2024
[68]

Bowen Peng, Jeﬀrey Quesnelle, and Diederik P. Kingma. 2 024. DeMo: Decoupled Momentum Optimization. arXiv: 2411.19870 [cs.LG] https://arxiv.org/abs/2411.19870
[69]

Kelsey Piper. 2024. Is AI progress slowing down? https://www.vox.com/future-perfect/389997/artiﬁcial-intelligence-openai-google-microsoft-chatgpt-progr ess-scaling. Vox

2024
[70]

Hadrien Pouget, Claire Dennis, Jon Bateman, Robert F. T rager, Renan Araujo, Haydn Belﬁeld, Belinda Cleeland, Malo u Estier, Gideon Futerman, Oliver Guest, Carlos Ignacio Guti errez, Vishnu Kannan, Casey Mahoney, Matthijs Maas, Charle s Martinet, Jakob Mökander, Kwan Yee Ng, S...

2024
[71]

Federal Register. 2023. Executive Order 14110 on the Sa fe, Secure, and Trustworthy Development and Use of Artiﬁcia l Intelligence. https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-developmen t-and-use-of-artiﬁcial-intelligence

2023
[72]

Kochenderfer

Anka Reuel, Amelia Hardy, Chandler Smith, Max Lamparth , Malcolm Hardy, and Mykel J. Kochenderfer. 2024. BetterBen ch: Assessing AI Bench- marks, Uncovering Issues, and Establishing Best Practices . arXiv: 2411.12990 [cs.AI] https://arxiv.org/abs/2411.12990

2024 arXiv
[73]

Kylie Robinson. 2024. OpenAI cofounder Ilya Sutskever says the way AI is built is about to change. https://www.theverge.com/2024/12/13/24320811/what-ilya-sutskever-sees-openai-model-data-training . The Verge

2024
[74]

Kylie Robinson. 2024. OpenAI teases new reasoning mode l—but don’t expect to try it soon.https://www.theverge.com/2024/12/20/24326036/openai-o1-o2-o3-reasoning-model-testing . The Verge

2024
[75]

Emma Roth. 2024. Google reveals AI ‘reasoning’ model th at ‘explicitly shows its thoughts’.https://www.theverge.com/2024/12/19/24325169/google-ai-reasoning-model-gemini-ﬂash-2-thinking The Verge

2024
[76]

Hadﬁeld, Richard Ngo, Konstantin Pilz, George Gor, Emma Bluemke, Sarah Shoker, Ja net Egan, Robert F

Girish Sastry, Lennart Heim, Haydn Belﬁeld, Markus And erljung, Miles Brundage, Julian Hazell, Cullen O’Keefe, Gi llian K. Hadﬁeld, Richard Ngo, Konstantin Pilz, George Gor, Emma Bluemke, Sarah Shoker, Ja net Egan, Robert F. Trager, Shahar Avin, Adrian Weller, Yosh ua Bengio, ...

2024 arXiv
[77]

OECD Secretariat. 2024. Public consultation on risk th resholds for advanced AI systems.https://oecd.ai/en/wonk/seeking-your-views-public-c onsultation-on-risk-thresholds-for-advanced-

2024
[78]

Deepa Seetharaman. 2024. The Next Great Leap in AI Is Beh ind Schedule and Crazy Expensive. https://www.wsj.com/tech/ai/openai-gpt5-orion-delay s-639e7693. The Wall Street Journal

2024
[79]

Matt Sheehan. 2024. China’s Views on AI Safety Are Chang ing—Quickly. https://carnegieendowment.org/research/2024/08/china-artiﬁcial-intelligence-ai-safety-regulation?lang= en. Carnegie Endowment for International Peace

2024
[80]

Matt Sheehan. 2024. Tracing the Roots of China’s AI Regu lations. https://carnegieendowment.org/research/2024/02/tracing-the-roots-of-chinas-ai-regulations?lang=en . Carnegie Endowment for International Peace

2024
[81]

David Shepardson. 2025. Trump revokes Biden executive order on addressing AI risks.https://www.reuters.com/technology/artiﬁcial-intelligence/trump-revokes-biden-executive-order-addressi ng- Reuters

2025
[82]

Ilya Sutskever. 2024. Sequence to sequence learning wi th neural networks: what a decade. NeurIPS Test of Time Talk

2024
[83]

Rich Sutton. 2019. The Bitter Lesson. http://www.incompleteideas.net/IncIdeas/BitterLesson.html. Incomplete Ideas

2019
[84]

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko reit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In Advances in Neural Information Processing Systems , I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S....

2017
[85]

Pablo Villalobos, Anson Ho, Jaime Sevilla, Tamay Besir oglu, Lennart Heim, and Marius Hobbhahn. 2024. Will we run ou t of data? Limits of LLM scaling based on human-generated data. arXiv: 2211.04325 [cs.LG] https://arxiv.org/abs/2211.04325

2024 arXiv
[86]

Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeﬀ Dean, and William Fedus

Jason Wei, Yi Tay, Rishi Bommasani, Colin Raﬀel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeﬀ Dean, and William Fedus. 2022. Emergent Abilities of Lar ge Language Mode...

2022 arXiv
[87]

An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, C hang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Hua ng, et al. 2024. Qwen2 technical report. arXiv preprint arXiv:2407.10671 (2024). Received 22 January 2025 19

2024 arXiv
[2024]

arXiv:2403.20194 [cs.MM] https://arxiv.org/abs/2403.20194

ConvBench: A Multi-Turn Conversation Evaluation Ben chmark with Hierarchical Capability for Large Vision-Lang uage Models. arXiv:2403.20194 [cs.MM] https://arxiv.org/abs/2403.20194

arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

European Union AI Act. 2024. Article 51: Classiﬁcation o f General-Purpose AI Models as General-Purpose AI Models wi th Systemic Risk. https://artiﬁcialintelligenceact.eu/article/51/

2024

[2] [2]

European Union AI Act. 2024. Chapters III and V. https://artiﬁcialintelligenceact.eu/article/51/

2024

[3] [3]

Meta AI. 2024. Introducing Meta Llama 3: The most capable openly available LLM to date. https://ai.meta.com/blog/meta-llama-3/

2024

[4] [4]

UK AI Safety Institute. 2024. Inspect AI: Framework for Large Language Model Evaluations . https://github.com/UKGovernmentBEIS/inspect_ai

2024

[5] [5]

Markus Anderljung, Joslyn Barnhart, Anton Korinek, Jad e Leung, Cullen O’Keefe, Jess Whittlestone, Shahar Avin, Mi les Brundage, Justin Bullock, Duncan Cass-Beggs, Ben Chang, Tantum Collins, Tim Fist, Gil lian Hadﬁeld, Alan Hayes, Lewis Ho, Sara Hooker, Eric Horvit z, Noam Kolt, Jonas Schuett, Yonadav Shavit, Divya Siddarth, Robert Trager, an d Kevin Wolf...

arXiv 2023

[6] [6]

Gregory Arcuri and Sujai Shivakumar. 2022. Moore’s Law a nd Its Practical Implications.https://www.csis.org/analysis/moores-law-and-its-pr actical-implications. Center for Security and International Studies

2022

[7] [7]

Yasaman Bahri, Ethan Dyer, Jared Kaplan, Jaehoon Lee, an d Utkarsh Sharma. 2024. Explaining neural scaling laws. Proceedings of the National Academy of Sciences 121, 27 (2024), e2311878121. doi:10.1073/pnas.2311878121 arXiv:https://www.pnas.org/doi/pdf/10.1073/pnas.2311878121

doi:10.1073/pnas.2311878121 2024

[8] [8]

Arsheeya Bajwa and Krystal Hu. 2024. Amazon doubles down on AI startup Anthropic with another $4 bln. https://www.reuters.com/technology/artiﬁcial-intelligence/anthropic-receives-4-billion-investment-amaz on-makes-aws-oﬃcial-cloud-provider-2024-11-22 . Reuters

2024