Data and AI governance: Promoting equity, ethics, and fairness in large language models

Alok Abhishek; Lisa Erickson; Tushar Bandopadhyay

REVIEW 3 major objections 4 minor 38 references

Data and AI governance: Promoting equity, ethics, and fairness in large language models

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A lifecycle governance loop built on a bias test suite can reduce discrimination risk in deployed large language models.

desk verdict Readable abstract promises a lifecycle governance framework on top of the authors' BEATS benchmark, but the body is a garbled blob; what is visible is an asserted effectiveness claim with no data, which is not yet referee-worthy. read the letter →

arxiv 2508.03970 v1 pith:I6GKKO4O submitted 2025-08-05 cs.CL cs.AI

Alok Abhishek , Lisa Erickson , Tushar Bandopadhyay This is my paper

classification cs.CLcs.AI

keywords biasevaluationlargelanguagemodelsAIgovernancedatafairnessethicsfactualityproductionmonitoring

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that bias in large language models can be systematically governed rather than merely audited after the fact. It argues that a data and AI governance process running across the full model lifecycle—development, validation, production, and post-deployment—can quantify bias, ethics, fairness, and factuality gaps, and can act on them. The centerpiece is BEATS, the Bias Evaluation and Assessment Test Suite, used first as a pre-production benchmark and then as a continuous real-time monitor with guardrails on generated responses. If the approach works as claimed, organizations have a practical way to benchmark LLMs before launch, catch drift after launch, and reduce the risk of discriminatory or reputationally harmful outputs.

What carries the argument

The central object is BEATS, the Bias Evaluation and Assessment Test Suite for large language models—a measurement instrument that quantifies bias and fairness gaps and provides the signal that drives the governance loop. The loop itself is the mechanism: pre-production benchmarking sets a baseline, continuous real-time evaluation detects drift or emerging bias in live outputs, and guardrails intercept or correct problematic generated responses before they reach users.

What would settle it

Compare BEATS outcomes with an independent audit of a deployed LLM's real user-facing outputs: if a model scores well on BEATS and passes the guardrails yet still produces measurably disparate treatment toward a protected group in live use—for example, consistently lower-quality, more hostile, or more restrictive responses—then the claim that the lifecycle governance loop mitigates discrimination risk would fail.

Watch

Extended reading notes

Core claim

The paper's central claim is that fairness and safety for LLMs are lifecycle properties, not one-time checkpoints. It proposes that a single governance stack built on the BEATS suite can (1) rigorously benchmark models for bias, ethics, fairness, and factuality before production deployment, (2) support continuous real-time evaluation once deployed, and (3) proactively govern model-generated responses through guardrails. Adopting this lifecycle governance, the paper argues, reduces discrimination risk and protects against brand or reputational harm. The contribution is an extension: it takes an existing bias test suite and turns it into an end-to-end governance procedure for generative AI sys

Load-bearing premise

The load-bearing premise is that scores from the BEATS suite track the bias and fairness harms that actually matter in real-world LLM outputs, so that passing the benchmarks and applying the guardrails genuinely lowers discrimination risk rather than just improving test performance.

Editorial extensions

If this is right

Deploying teams can run BEATS as a pre-launch gate, benchmarking an LLM's bias, ethics, fairness, and factuality before production.
Post-deployment, the same suite can monitor outputs in real time and flag drift that static evaluations would miss.
Guardrails tied to the bias signal can proactively block or rewrite discriminatory responses, lowering discrimination and brand risk.
Applying one governance framework across the lifecycle makes safety and responsibility properties of the deployment process, not just of the model weights.
The framework extends bias testing from model evaluation to organizational data and AI governance.

Reading between the lines

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

If BEATS captures real-world fairness harm, a natural next step the paper leaves implicit is using it as a procurement or certification standard: vendors could be required to publish lifecycle bias scores before a contract is signed.
The same lifecycle loop is portable to multimodal and agentic systems, where bias can surface in tool selection and action sequences rather than only in generated text; this would be a testable extension of the suite.
Because the paper groups factuality with ethics and fairness under one governance stack, an unresolved question it does not address is whether a single score can serve both accuracy and equity without trading one off against the other.
A guardrail that blocks biased responses necessarily sets a threshold for what counts as biased; the paper implies such thresholds can be chosen objectively, but the choice itself is a policy decision.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

You asked about 2508.03970. Short version: I could not read the full text. The PDF text layer is corrupted, so any opinion has to rest on the abstract. That is the first and most concrete problem: a submission that cannot be read by a human or a machine is, in practical terms, not a submission. A desk reject makes sense until the authors fix the file.

What the abstract shows is a governance framework built on the authors' own BEATS bias-evaluation suite, covering the whole lifecycle from development to production monitoring and guardrails. The framing is sensible. Naming fairness, factuality, and ethics as continuous governance problems rather than one-time audits is a useful stance, and the field could use practical protocols for real-time evaluation. That part gets credit.

What is actually new is impossible to judge. The abstract presents this as an extension of prior BEATS work, but without the body I cannot see whether there is a new metric, a dataset, a case study, or a protocol. The central claim — that implementing this governance significantly enhances safety and mitigates discrimination risk — is asserted with no visible evidence. That is not fatal on its own if the full text contains rigorous evaluation, but the burden is real.

The conceptual soft spot is the self-referentiality. BEATS is both the measurement instrument and the foundation of the framework. That does not automatically make it circular, but without external validation against established bias benchmarks, human judgments, or real downstream harm, the framework risks giving false assurance. The abstract gives no hint that such validation exists. The stress-test note points to this as the load-bearing assumption, and I agree: it is exactly where the paper needs to deliver.

The audience for this paper is practitioners wanting a governance checklist, not researchers looking for new methods. If the body supplies the missing detail, it might be a useful engineering report. I would not cite it now, and I would not send it to peer review in this form. Fix the PDF, add an external validation of BEATS, and show at least one concrete deployment where the monitoring caught something. Then it becomes a different paper. As it stands, desk reject with an invitation to resubmit.

Referee Report

3 major / 4 minor

Summary. This paper proposes a data and AI governance framework, built on the authors' BEATS bias-evaluation suite, to govern bias, ethics, fairness, and factuality across the LLM lifecycle. The abstract claims the framework is suitable for practical real-world use, enables pre-deployment benchmarking and continuous evaluation, and that implementing it will significantly enhance safety/responsibility and mitigate discrimination risk. The only readable portion is the abstract; the supplied full text is a mis-encoded, uninterpretable character stream, so no technical claims, tables, equations, or results can be inspected.

Significance. If the framework worked as claimed, it would be valuable: operationalizing fairness governance across development, deployment, and monitoring is an important open problem. The paper, however, provides no visible evaluative evidence: no effect sizes, no comparisons, no external validation of BEATS, no reproducibility artifacts. The manuscript's central assertion therefore cannot be credited from the submitted text.

major comments (3)

[Full text (all sections after the abstract)] The full text is a garbled character stream that cannot be read as prose, equations, or data. As a result, the central claims in the abstract—'suitable for practical, real-world applications,' 'significantly enhance the safety and responsibility,' 'effectively mitigating risks of discrimination'—cannot be checked against any derivation, benchmark, or evaluation. This is a load-bearing problem, not a stylistic issue.
[Abstract, first sentence] The framework is grounded in 'our foundational work on the Bias Evaluation and Assessment Test Suite (BEATS).' The effectiveness of the governance approach depends on BEATS scores being a valid proxy for real-world fairness harm. No external validation is reported or visible: there is no comparison of BEATS with established bias benchmarks (e.g., BBQ, StereoSet), no correlation with human fairness judgments, and no downstream harm measure. Without such validation, a model passing BEATS may provide false assurance rather than reduced discrimination risk.
[Abstract, final sentences] The claim that the governance approach 'significantly enhance[s] the safety and responsibility' and 'effectively mitigat[es] risks of discrimination' is an empirical causal claim about intervention effectiveness. No data, controlled comparison, or deployment study is supplied in the readable text. If the full text contains such evidence, this comment should be revisited; as submitted, the evidence is absent.

minor comments (4)

[Abstract] The phrase 'the authors share' should be 'we share' for consistency with a single-author paper.
[Abstract] The acronym 'GenAI' is used without expansion; spell out 'generative artificial intelligence' at first use or define the abbreviation.
[Full text] The submission is corrupted. If this is an encoding artifact, the authors must resubmit a readable PDF or TeX source with all Unicode intact; otherwise the manuscript cannot be processed.
[Abstract] Capitalization of 'Bias, Ethics, Fairness, and Factuality' should be lower-case unless these are proper nouns or defined category names.

Circularity Check

1 steps flagged · score 4.0 of 10

Self-citation to the authors' own BEATS suite is load-bearing; no formal derivation is visible because the body is corrupted, but the fairness/benchmarking claim reduces to an internally defined measure.

self citation load bearing [Abstract, first two sentences and governance-claim sentence]
"Building upon our foundational work on the Bias Evaluation and Assessment Test Suite (BEATS) for Large Language Models, the authors share prevalent bias and fairness related gaps in Large Language Models (LLMs) and discuss data and AI governance framework to address Bias, Ethics, Fairness, and Factuality within LLMs. ... The data and AI governance approach discussed in this paper is suitable for practical, real-world applications, enabling rigorous benchmarking of LLMs prior to production deployment, facilitating continuous real-time evaluation, and proactively governing LLM generated response"

The abstract gives no independent criterion for 'rigorous benchmarking' or for 'mitigating risks of discrimination'; it explicitly grounds the governance framework in the authors' own BEATS suite. Thus the paper's headline claims of enabling rigorous benchmarking and reducing discrimination rest on a measurement instrument created by the same authors, with no external validation against established bias benchmarks, human judgments, or real-world harm measures visible in the readable text. This is a load-bearing self-citation: the framework's effectiveness is assessed with the authors' own metric, so the claim of suitability is, at least in part, defined by the self-cited instrument. The corrupted full text prevents checking whether any independent validation is supplied later, but on the a

full rationale

The only mechanically readable portion of this submission is the abstract; the body is a corrupted byte sequence (mojibake) with no recoverable equations, tables, or section text. In the abstract, the paper's central governance claim is explicitly grounded in the authors' own prior BEATS suite, and the suitability claim is not tied to any external benchmark or independent outcome measure. If BEATS is the instrument by which 'rigorous benchmarking' and 'mitigating risks of discrimination' are assessed, then the framework's headline effectiveness partially reduces to a self-cited, internally defined measure. I do not call this a full formal circularity because no equation or fitted parameter can be inspected in the corrupted full text, and the governance-lifecycle discussion has independent content beyond BEATS. The score of 4 reflects the load-bearing self-citation and the unverifiable body, not a demonstrated 'prediction = input' construction; it is a measurement-validity dependence rather than an explicit derivation collapse. Self-citation alone would not justify a higher score, and no appended limitation statement or omitted-proof note is readable to weigh further.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

Abstract-only ledger. No fitted numbers, thresholds, or invented entities are visible in the readable text; the framework rests on domain assumptions about the validity of bias benchmarks (BEATS), the transfer of benchmark scores to real-world harm, and the causal effectiveness of lifecycle governance. Any thresholds embedded in the full framework would be free parameters, but they are not stated in the abstract.

assumptions (3)

domain assumption Quantitative bias benchmarks are valid proxies for real-world fairness harm in LLM systems
The governance approach assumes that scores on test suites (BEATS) correspond to actual discrimination risk in production. Invoked throughout the abstract, e.g., 'enabling rigorous benchmarking of LLMs prior to production deployment'.
domain assumption The authors' prior BEATS suite is a sound, reliable measurement instrument for LLM bias
Abstract: 'Building upon our foundational work on the Bias Evaluation and Assessment Test Suite (BEATS)'. The present claims inherit the validity of the prior suite; no external validation is cited in the abstract.
domain assumption Lifecycle governance interventions (monitoring, guardrails, continuous evaluation) causally reduce bias-related harm once deployed
Abstract: 'By implementing the data and AI governance across the life cycle of AI development, organizations can significantly enhance the safety and responsibility of their GenAI systems.' This causal effectiveness is assumed.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Data and AI governance: Promoting equity, ethics, and fairness in large language models." pith.science (2026). https://pith.science/paper/I6GKKO4O

@misc{pith2026250803970,
  author       = {Pith},
  title        = {Pith review of: Data and AI governance: Promoting equity, ethics, and fairness in large language models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I6GKKO4O}},
  note         = {Machine review of arXiv:2508.03970}
}

read the original abstract

In this paper, we cover approaches to systematically govern, assess and quantify bias across the complete life cycle of machine learning models, from initial development and validation to ongoing production monitoring and guardrail implementation. Building upon our foundational work on the Bias Evaluation and Assessment Test Suite (BEATS) for Large Language Models, the authors share prevalent bias and fairness related gaps in Large Language Models (LLMs) and discuss data and AI governance framework to address Bias, Ethics, Fairness, and Factuality within LLMs. The data and AI governance approach discussed in this paper is suitable for practical, real-world applications, enabling rigorous benchmarking of LLMs prior to production deployment, facilitating continuous real-time evaluation, and proactively governing LLM generated responses. By implementing the data and AI governance across the life cycle of AI development, organizations can significantly enhance the safety and responsibility of their GenAI systems, effectively mitigating risks of discrimination and protecting against potential reputational or brand-related harm. Ultimately, through this article, we aim to contribute to advancement of the creation and deployment of socially responsible and ethically aligned generative artificial intelligence powered applications.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

38 extracted references · 23 canonical work pages

[1]

11em plus .33em minus .07em @technote 4000 4000 100 4000 4000 500 `\.=1000 = #1 #1 #1 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEauthorblockAstyle \@IEEEauthordefaulttextstyle \@IEEEauthorblockconfadjspace -0.25em \@IEEEauthorblockNtopspace 0.0ex \@IEEEauthorblockAtopspace 0.0ex \@IEEEauthorblockNinterlinespace 2.6ex \@IEEEauthorblockAinte...

work page
[2]

title Generative AI to become a \ 1.3 trillion market by 2032, research finds ( year 2023 )

author Bloomberg . title Generative AI to become a \ 1.3 trillion market by 2032, research finds ( year 2023 ). note Online: https://www.bloomberg.com/company/press/generative-ai-to-become-a-1-3-trillion-market-by-2032-research-finds/

work page
[3]

BEATS: Bias Evaluation and Assessment Test Suite for Large Language Models

author Abhishek, A. , author Erickson, L. & author Bandopadhyay, T. title BEATS : Bias evaluation and assessment test suite for large language models . journal arXiv ( year 2025 ). note https://arxiv.org/abs/2503.24310

work page Pith review arXiv 2025
[4]

title Worldwide spending on artificial intelligence forecast to reach \ 632 billion in 2028 ( year 2023 )

author International Data Corporation (IDC) . title Worldwide spending on artificial intelligence forecast to reach \ 632 billion in 2028 ( year 2023 ). note Online: https://www.idc.com/getdoc.jsp?containerId=prUS52530724

work page
[5]

title Generative AI market forecasts revised upward to \ 52.2b by 2028 ( year 2023 )

author S&P Global . title Generative AI market forecasts revised upward to \ 52.2b by 2028 ( year 2023 ). note Online: https://www.spglobal.com/marketintelligence/en/news-insights/research/generative-ai-market-forecasts-revised-upward-to-52-2b-by-2028

work page
[6]

title Generative AI spending to reach \ 26 billion by 2027 ( year 2023 )

author International Data Corporation (IDC) . title Generative AI spending to reach \ 26 billion by 2027 ( year 2023 ). note Online: https://www.idc.com/getdoc.jsp?containerId=prAP52048824

work page 2027
[7]

title Generative AI market size, share, and trends 2024 to 2033 ( year 2023 )

author Precedence Research . title Generative AI market size, share, and trends 2024 to 2033 ( year 2023 ). note Online: https://www.precedenceresearch.com/generative-ai-market

work page 2024
[8]

title Spend on generative AI will grow 36\ note Online: https://www.forrester.com/blogs/spend-on-generative-ai-will-grow-36-annually-to-2030/

author Forrester . title Spend on generative AI will grow 36\ note Online: https://www.forrester.com/blogs/spend-on-generative-ai-will-grow-36-annually-to-2030/

work page

Show all 38 references

[9]

, author Hu, Q

author Norori, N. , author Hu, Q. , author Aellen, F. M. , author Faraci, F. D. & author Tzovara, A. title Addressing bias in big data and AI for health care: A call for open science . journal Patterns (N Y) volume 2 , pages 100347 ( year 2021 ). note https://www.ncbi.nlm.nih....

2021
[10]

& author Alhashmi, S

author Alhosani, K. & author Alhashmi, S. M. title Opportunities, challenges, and benefits of ai innovation in government services: a review . journal Discover Artificial Intelligence volume 4 ( year 2024 ). note https://link.springer.com/article/10.1007/s44163-024-00111-w

2024 doi
[11]

, author Chang, K.-W

author Bolukbasi, T. , author Chang, K.-W. , author Zou, J. , author Saligrama, V. & author Kalai, A. title Man is to computer programmer as woman is to homemaker? debiasing word embeddings . journal arXiv ( year 2016 ). note https://arxiv.org/abs/1607.06520

2016 arXiv
[12]

, author Tan, I

author Luna, J. , author Tan, I. , author Xie, X. & author Jiang, L. title Navigating governance paradigms: A cross-regional comparative study of generative AI governance processes & principles . In booktitle Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , ...

2024 doi
[13]

author de Almeida, P. G. R. , author dos Santos, C. D. & author Farias, J. S. title Artificial intelligence regulation: a framework for governance . journal Ethics and Information Technology volume 23 ( year 2021 ). note https://link.springer.com/article/10.1007/s10676-021-09593-z

2021 doi
[14]

, author Cai, X

author Li, J. , author Cai, X. & author Cheng, L. title Legal regulation of generative AI : a multidimensional construction . journal International Journal of Legal Discourse volume 8 , pages 365--388 ( year 2023 ). note https://www.degruyterbrill.com/document/doi/10.1515/ijld...

2023 doi
[16]

author Vuković, D. B. , author Dekpo-Adza, S. & author Matović, S. title Ai integration in financial services: a systematic review of trends and regulatory challenges . journal Humanities and Social Sciences Communications volume 12 ( year 2025 ). note https://link.springer.co...

2025 doi
[17]

, author Lin, E

author Palaniappan, K. , author Lin, E. Y. T. & author Vogel, S. title Global regulatory frameworks for the use of artificial intelligence ( AI ) in the healthcare services sector . In booktitle Healthcare , vol. volume 12 , pages 562 ( year 2024 ). note https://www.mdpi.com/2...

2024
[18]

, author Henderson, P

author Kapoor, S. , author Henderson, P. & author Narayanan, A. title Promises and pitfalls of artificial intelligence for legal applications ( year 2024 ). note https://arxiv.org/abs/2402.01656 , 2402.01656

2024 arXiv
[19]

author Magesh, V. et al. title Hallucination-free? assessing the reliability of leading ai legal research tools ( year 2024 ). note https://arxiv.org/abs/2405.20362 , 2405.20362

2024 arXiv
[20]

, author Janssen, M

author Mellouli, S. , author Janssen, M. & author Ojo, A. title Introduction to the issue on artificial intelligence in the public sector: Risks and benefits of ai for governments . journal Digit. Gov.: Res. Pract. volume 5 ( year 2024 ). note https://dl.acm.org/doi/full/10.11...

2024 doi
[21]

, author Agdas, D

author Yigitcanlar, T. , author Agdas, D. & author Degirmenci, K. title Artificial intelligence in local governments: perceptions of city managers on prospects, constraints and choices . journal AI & SOCIETY volume 38 ( year 2023 ). note https://link.springer.com/article/10.10...

2023 doi
[22]

author Parrish, A. et al. title BBQ : A hand-built bias benchmark for question answering . journal Findings of the Association for Computational Linguistics: ACL 2022 volume 2022 ( year 2022 ). note https://aclanthology.org/2022.findings-acl.165/

2022
[23]

title The Basel Committee on Banking Supervision: A History of the Early Years, 1974--1997 ( publisher Cambridge University Press , year 2011 )

author Goodhart, C. title The Basel Committee on Banking Supervision: A History of the Early Years, 1974--1997 ( publisher Cambridge University Press , year 2011 ). note https://doi.org/10.1017/CBO9780511996238

1974 doi
[24]

title Iso 14001 - environmental management systems — requirements with guidance for use

author International Organization for Standardization . title Iso 14001 - environmental management systems — requirements with guidance for use . howpublished https://www.iso.org/standard/60857.html ( year 2015 )

2015
[25]

author Slattery, P. et al. title The AI risk repository: A comprehensive meta-review, database, and taxonomy of risks from artificial intelligence . journal arXiv ( year 2025 ). note https://arxiv.org/abs/2408.12622

2025 arXiv
[26]

author Weidinger, L. et al. title Ethical and social risks of harm from language models ( year 2021 ). note https://arxiv.org/abs/2112.04359

2021 arXiv
[27]

, author Al-Mallah, M

author Elshawi, R. , author Al-Mallah, M. H. & author Sakr, S. title On the interpretability of machine learning-based model for predicting hypertension . journal BMC Med. Inform. Decis. Mak. volume 19 , pages 146 ( year 2019 ). note https://www.ncbi.nlm.nih.gov/pmc/articles/P...

2019
[28]

author Qureshi, N. I. , author Choudhuri, S. S. , author Yaramala Nagamani, R. A. V. & author Shah, R. title Ethical considerations of ai in financial services: Privacy, bias, and algorithmic transparency . journal 2024 International Conference on Knowledge Engineering and Com...

2024
[29]

& author Zhou, L

author Zhang, Y. & author Zhou, L. title Fairness assessment for artificial intelligence in financial industry ( year 2019 ). note https://arxiv.org/abs/1912.07211 , 1912.07211

2019 arXiv
[30]

author Silva, D. D. & author Alahakoon, D. title An artificial intelligence life cycle: From conception to production . journal Patterns volume 3 ( year 2022 ). note https://doi.org/10.1016/j.patter.2022.100489

2022
[31]

, author Cruz, L

author Haakman, M. , author Cruz, L. , author Huijgens, H. & author van Deursen, A. title Ai lifecycle models need to be revised. an exploratory study in fintech . journal Empirical Software Engineering volume 26 ( year 2022 ). note https://doi.org/10.1007/s10664-021-09993-1

2022 doi
[32]

title Explainable artificial intelligence ( year 2017 )

author Wikipedia . title Explainable artificial intelligence ( year 2017 ). note Online (Accessed 12-Apr-2025): https://en.wikipedia.org/wiki/Explainable_artificial_intelligence

2017
[33]

& author Lee, S.-I

author Lundberg, S. & author Lee, S.-I. title A unified approach to interpreting model predictions . journal arXiv ( year 2017 ). note https://arxiv.org/abs/1705.07874

2017 arXiv
[34]

W hy should i trust you?

author Ribeiro, M. T. , author Singh, S. & author Guestrin, C. title " W hy should i trust you?": Explaining the predictions of any classifier . journal arXiv ( year 2016 ). note https://arxiv.org/abs/1602.04938

2016 arXiv
[35]

, author Herbinger, J

author Moosbauer, J. , author Herbinger, J. , author Casalicchio, G. , author Lindauer, M. & author Bischl, B. title Explaining hyperparameter optimization via partial dependence plots . journal arXiv ( year 2022 ). note https://arxiv.org/abs/2111.04820

2022 arXiv
[36]

, author Dickerson, J

author Verma, S. , author Dickerson, J. & author Hines, K. title Counterfactual explanations for machine learning: Challenges revisited . journal arXiv ( year 2021 ). note https://arxiv.org/abs/2106.07756

2021 arXiv
[37]

author Ye, J. et al. title Justice or prejudice? Q uantifying biases in LLM -as-a-judge . journal arXiv ( year 2024 ). note https://arxiv.org/abs/2410.02736

2024 arXiv
[38]

author Zheng, L. et al. title Judging LLM -as-a-judge with MT-Bench and Chatbot Arena . journal arXiv ( year 2023 ). note https://arxiv.org/abs/2306.05685

2023 arXiv
[39]

author Vaswani, A. et al. title Attention is all you need . journal arXiv ( year 2023 ). note https://arxiv.org/abs/1706.03762

2023 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

11em plus .33em minus .07em @technote 4000 4000 100 4000 4000 500 `\.=1000 = #1 #1 #1 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEauthorblockAstyle \@IEEEauthordefaulttextstyle \@IEEEauthorblockconfadjspace -0.25em \@IEEEauthorblockNtopspace 0.0ex \@IEEEauthorblockAtopspace 0.0ex \@IEEEauthorblockNinterlinespace 2.6ex \@IEEEauthorblockAinte...

work page

[2] [2]

title Generative AI to become a \ 1.3 trillion market by 2032, research finds ( year 2023 )

author Bloomberg . title Generative AI to become a \ 1.3 trillion market by 2032, research finds ( year 2023 ). note Online: https://www.bloomberg.com/company/press/generative-ai-to-become-a-1-3-trillion-market-by-2032-research-finds/

work page

[3] [3]

BEATS: Bias Evaluation and Assessment Test Suite for Large Language Models

author Abhishek, A. , author Erickson, L. & author Bandopadhyay, T. title BEATS : Bias evaluation and assessment test suite for large language models . journal arXiv ( year 2025 ). note https://arxiv.org/abs/2503.24310

work page Pith review arXiv 2025

[4] [4]

title Worldwide spending on artificial intelligence forecast to reach \ 632 billion in 2028 ( year 2023 )

author International Data Corporation (IDC) . title Worldwide spending on artificial intelligence forecast to reach \ 632 billion in 2028 ( year 2023 ). note Online: https://www.idc.com/getdoc.jsp?containerId=prUS52530724

work page

[5] [5]

title Generative AI market forecasts revised upward to \ 52.2b by 2028 ( year 2023 )

author S&P Global . title Generative AI market forecasts revised upward to \ 52.2b by 2028 ( year 2023 ). note Online: https://www.spglobal.com/marketintelligence/en/news-insights/research/generative-ai-market-forecasts-revised-upward-to-52-2b-by-2028

work page

[6] [6]

title Generative AI spending to reach \ 26 billion by 2027 ( year 2023 )

author International Data Corporation (IDC) . title Generative AI spending to reach \ 26 billion by 2027 ( year 2023 ). note Online: https://www.idc.com/getdoc.jsp?containerId=prAP52048824

work page 2027

[7] [7]

title Generative AI market size, share, and trends 2024 to 2033 ( year 2023 )

author Precedence Research . title Generative AI market size, share, and trends 2024 to 2033 ( year 2023 ). note Online: https://www.precedenceresearch.com/generative-ai-market

work page 2024

[8] [8]

title Spend on generative AI will grow 36\ note Online: https://www.forrester.com/blogs/spend-on-generative-ai-will-grow-36-annually-to-2030/

author Forrester . title Spend on generative AI will grow 36\ note Online: https://www.forrester.com/blogs/spend-on-generative-ai-will-grow-36-annually-to-2030/

work page