Pith. sign in

REVIEW 4 major objections 4 minor 61 references

Freemium Is All You Need

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

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

Pith's one-line read For uniform user values, the profit-maximizing freemium strategy for a generative-AI service has a four-case closed form.

desk verdict The steady-state analysis is solid, but the paper does not solve its claimed dynamic problem, and a sign error in Eq. (9) breaks the closed-form case classification. read the letter →

arxiv 2608.00823 v2 pith:BV5FNZ7A submitted 2026-08-01 cs.GT

Maxim Bichuch , Aviv Yaish This is my paper · ORCID
classification cs.GT
keywords freemiumpricinggenerativeAIplatformeconomicsprivacy-sensitivedemandqualitydynamicssteadystatevaluethresholdsclosed-formoptimization
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 asks when a generative-AI provider should offer a free tier whose requests improve the model, given that paid requests stay private and every request costs compute. It argues that, for a menu of one free training-eligible tier and one paid training-exempt tier, the entire optimization can be re-expressed through two endogenous value thresholds that separate privacy-sensitive paid users from public users. For user values uniformly distributed on [0,M], it derives a closed-form optimal strategy with four cases and shows that a positive-quality free tier is optimal exactly in the interior and boundary cases where the data-learning engine outweighs compute costs. If right, the result turns the freemium decision into a direct comparison of inference cost, learning and decay rates, and the share of privacy-sensitive demand.

What carries the argument

The argument runs on two value thresholds, $v_1=c/Q$ for privacy-sensitive paid demand and $v_2=c/(Q(1-\gamma))$ for public demand, derived from the demand characterization in Proposition 6.2. A change of variables from the controls $(c,\gamma)$ to $(v_1,v_2)$ decouples monetization from quality: revenue and cost become functions of the two thresholds alone. The Nerlove-Arrow quality equation $\dot Q=\alpha N_\phi-\delta Q$ provides the steady-state identity $Q=K F(v_2)$, and the uniform distribution reduces profit maximization to a concave program over two normalized thresholds whose first-order conditions and Hessian sign deliver the four cases.

What would settle it

Solve numerically the true dynamic program of Definition 5.1 with $\dot Q=\alpha N_\phi-\delta Q$ for uniformly distributed values and compare its optimal trajectory's discounted profit with the steady-state profit from Proposition 10.8; a parameter region where the dynamic policy strictly outperforms the stationary four-case policy would show that the steady-state reduction is not equivalent to the original problem.

Watch

Extended reading notes

Core claim

The central claim is that freemium is the optimal long-run strategy precisely when the quality-generation potential $K = \alpha\lambda(1-\varphi)/\delta$ is strong relative to the expected compute cost generated by privacy-sensitive users, $C\varphi$. With uniformly distributed values on $[0,M]$, the optimal thresholds $x_1=v_1/M$ and $x_2=v_2/M$ are given in closed form by Proposition 10.8: when the unconstrained maximizer is feasible, $x_1^*=\frac{2-\sqrt{1-\frac{12C\varphi(1-\varphi)}{KM}}}{3\varphi}$ and $x_2^*=\frac{1-\varphi x_1^*}{2(1-\varphi)}$; when the first constraint fails, both thresholds collapse to $\frac{1+\sqrt{1+\frac{3C\varphi}{KM}}}{3}$ with $\gamma=0$; when the second fails at moderate cost, $x_2^*=1$ and $x_1$ stays at its interior value; and when both fail, $x_1^*=x_2^*=1$ with $\gamma=0$. A positive-quality free tier, $\gamma>0$, is therefore optimal exactly in the first and third of these four cases.

Load-bearing premise

The load-bearing premise is that maximizing discounted long-run profit is equivalent to maximizing instantaneous steady-state profit, even though the model's own quality equation ties current free-tier demand to future quality and the paper asserts rather than proves that equivalence.

Editorial extensions

If this is right

  • For uniform user values, a provider can read off the optimal menu from the ratio $C\varphi/(KM)$ by comparing it with the thresholds in Equations (8) and (9).
  • When the ratio is high enough, the optimal strategy collapses to $\gamma=0$, a single paid tier with no training-eligible free tier.
  • When the ratio is moderate and the public segment is large, the provider optimally sets $x_2=1$, pushing all public users onto the free tier and monetizing only privacy-sensitive users.
  • Every stationary privacy-separated menu has a unique steady-state quality, reached globally from any initial quality, so the four cases describe a stable long-run outcome rather than a transitory one.
  • The free tier never starves: steady-state free-tier demand is strictly positive in the privacy-separated regime, so a training-eligible tier, when offered, always generates some training data.

Reading between the lines

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

  • [Inference] The closed-form comparison suggests a direct managerial rule: treat $C\varphi/(KM)$ as a single statistic; below the lower threshold run a freemium two-tier menu, above it collapse to paid-only or full-data-harvesting policies.
  • [Inference] Although the closed form is for uniform values, the two-threshold reduction itself is distribution-free; a provider with any continuous value distribution could run the same $(v_1,v_2)$ maximization numerically, with freemium optimal whenever the interior point satisfies the feasibility constraints.
  • [Inference] Because the paper's dynamic-to-steady-state equivalence is asserted rather than proved, the four cases are best read as the steady-state optimum; a full dynamic treatment could show where time-varying pricing beats the stationary policy when quality accumulation matters during the approach path.
  • [Inference] The model implies a sharp, testable cross-sectional prediction: services with higher inference cost per request or larger privacy-sensitive shares should be less likely to offer training-eligible free tiers, other things equal.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper studies a generative-AI provider that offers a paid, privacy-preserving tier and a free tier whose requests can be used for model training. User types differ in request value and privacy preference, quality evolves according to Qdot = alpha*N_phi - delta*Q, and the provider maximizes discounted profit by choosing price and the free tier's relative quality. The paper characterizes deterministic menus (Proposition 4.6, Corollary 4.8), derives demand thresholds (Proposition 6.2), proves existence, uniqueness, and global convergence of a steady state for fixed stationary strategies (Propositions 7.2-7.6), and then claims a closed-form solution of the dynamic optimization problem for uniformly distributed values (Proposition 10.8), with conditions under which a positive-quality free tier is optimal. Appendices contain full proofs.

Significance. The paper addresses a timely and economically interesting question: when should a GenAI provider offer a free, training-eligible tier even though free requests cost compute. The steady-state analysis of quality dynamics is self-contained and correctly proves, for a fixed stationary strategy, existence, uniqueness, and global convergence of quality; the demand characterization with value thresholds is also clearly derived. The paper does not rely on fitted parameters and makes an explicit falsifiable prediction in Proposition 10.8. However, the central claim that Proposition 10.8 solves the dynamic problem of Definition 5.1 is not established, and there are algebraic and boundary-condition errors in the closed-form result. The strengths are real but they concern a static steady-state problem, not the dynamic optimum claimed in the title and abstract.

major comments (4)
  1. [8, Definition 5.1] The paper asserts in Section 8 that maximizing Definition 5.1 is equivalent to maximizing Eq. (2) at all tau, 'under the assumption of no dynamic state equations,' and says Propositions 7.3 and 7.4 justify this. Those propositions only show that for a fixed stationary strategy (c,gamma), a unique steady state exists and is globally attracting; they say nothing about the optimality of a stationary policy in a discounted optimal control problem with state equation (1). Since the current control gamma(t) affects future quality through N_phi(t) in Eq. (1), the steady-state profit objective in Eq. (5) is not the same as the integral objective in Definition 5.1. No Pontryagin necessary conditions or dynamic-programming argument is given. A concrete check would be to compare the paper's steady-state policy against a two-phase policy that initially subsidizes the free tier to build quality and then monetizes; the paper provides no argument that the former dominates. Proposition 10.8 therefore does not solve the stated dynamic problem.
  2. [Eq. (9), Proposition 10.6] The inequality in Eq. (9) is algebraically reversed. From the proof of Proposition 10.6, x_2^* <= 1 is equivalent to phi x_1^* >= 2phi - 1, which, using x_1^* = (2 - sqrt(1 - 12 C phi (1-phi)/(KM)))/(3phi), gives after squaring C phi/(KM) >= 3phi - 2, not C phi/(KM) <= 3phi - 2. This flips the case classification in Proposition 10.8. For example, with phi = 0.8 and C/KM = 0.1, the printed Eq. (9) is satisfied but the correct condition fails (x_2^* > 1), so the 'unconstrained' Case 1 is infeasible and the claimed boundary behavior is wrong.
  3. [9, Eq. (6)] The demand formulas in Proposition 6.2 and Corollary 6.3 are derived under the privacy-separation condition, which by Proposition 6.1 holds iff L >= c gamma. In the reformulated variables of Section 9, c gamma = K M x_1 (x_2 - x_1). The optimization in Eqs. (5)-(6) imposes only 0 < x_1 <= x_2 <= M (or x_2 <= 1 after scaling); it never imposes L >= K M x_1 (x_2 - x_1). Thus the objective maximized in Propositions 10.5-10.8 may use demand functions that are invalid for the chosen controls. The paper must either add this constraint or prove that it is slack at the optimum for a stated range of L; as written, Proposition 10.8 is conditional on an unstated feasibility condition.
  4. [Proposition 10.8, Cases 3 and 4] When a variable is fixed at a bound, the first-order conditions must be re-derived. In Case 3, x_2 is set to 1 but x_1 is still taken from the unconstrained Proposition 10.5. The correct maximizer of J(x_1,1) solves partial J/partial x_1 = KM phi (1 - 2x_1) + C phi = 0, which gives x_1 = (1 + C/KM)/2, not the unconstrained formula. For phi = 0.8 and C/KM = 0.1, the paper's Case 3 solution yields J approximately 0.135 KM, while the boundary-optimal x_1 = 0.55 yields J approximately 0.142 KM. Case 4 is also not the constrained optimum: with phi = 0.8 and C/KM = 0.5, the Case 4 condition holds, but x_2 = 1, x_1 = (1 + C/KM)/2 = 0.75 yields a higher profit than the claimed x_1 = x_2 = 1. The boundary cases must be re-solved from the Kuhn-Tucker conditions.
minor comments (4)
  1. [Section 5, Eq. (1)] The notation N_phi(tau) in the quality dynamics is used before demand is defined in Section 6; either define demand first or refer the reader forward when the equation is introduced.
  2. [Proposition 10.2] The proof begins by citing Proposition 6.1's condition c gamma <= L, but the derived bound c/(1-gamma) <= M alpha lambda (1-phi)/delta does not involve L. Clarify whether L plays any role in the derivation or state that the bound is independent of L.
  3. [General] There are several typos and LaTeX artifacts: 'quaity' in the proof sketch of Proposition 7.6, 'weather' in the proof of Proposition 10.8, and the unresolved control sequence \gls{PDF} in reference [11]. Figure 1 is referenced but not included in the provided text; ensure it appears in the final version.
  4. [Definition 5.1] The admissible control space for c(tau) and gamma(tau) is not stated formally; add measurability and boundedness conditions so that the discount integral is well-defined and the maximization is over a precise set.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the closed-form thresholds derive from the model's own stated dynamics and profit objective, not from fitted inputs or load-bearing self-citation.

full rationale

The derivation chain is self-contained. Demand thresholds v1=c/Q and v2=c/(Q(1-gamma)) are definitions from the model (Prop. 6.2); steady-state quality Q=alpha/delta*N_phi follows by setting Qdot=0 in Eq. (1) (Prop. 7.2); the change of variables Q(v2)=K F(v2) and c=v1 K F(v2) is algebraic from those definitions; the objective Eq. (5) is the steady-state profit 1/r*(Rev-Cost) written in these variables; and the uniform-distribution closed form in Props. 10.5 and 10.8 comes from first-order conditions and constraint checking on that objective. No parameter is fitted to a target quantity, and no 'prediction' is inserted as an input. The only self-citation (Ref. [50], Wang and Yaish) appears in Related Work and is not load-bearing. The Section 8 move from the dynamic Definition 5.1 to pointwise/steady-state maximization is an unsupported optimality-scope reduction, but it is not a circular reduction: Proposition 10.8 solves the static problem the paper actually writes down rather than assuming that solution. Concerns about the dynamic equivalence and the dropped L>=c*gamma constraint belong to correctness, not circularity.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The conclusions rest on a specific dynamic model and several behavioral constraints (no screening, privacy separation, tie-breaking) that are assumed rather than derived. The algebraic derivation of the closed-form solution, if corrected, would still be conditional on these assumptions and on the uniform distribution.

assumptions (6)
  • domain assumption Quality evolves as Qdot = alpha*N_phi - delta*Q (Nerlove-Arrow form)
    Adopted in Section 5, Eq. (1), following [45,7]; it is a modeling choice, not derived from primitives.
  • domain assumption Uniform value distribution over [0,M] for the closed-form results
    Section 10 assumes f(v)=1/M on [0,M]; results are specialized to this case.
  • ad hoc to paper No screening: arbitrarily low-value non-private requests can use the training-eligible tier
    Definition 4.5; this behavioral constraint drives the free-tier necessity result.
  • ad hoc to paper Privacy separation: privacy-sensitive requests only choose training-exempt allocation or no service
    Definition 4.7; needed for the demand formulas in Prop 6.2, and requires L>=c*gamma.
  • ad hoc to paper Tie-breaking: indifferent users choose the free tier, so N_phi is positive even when gamma=0
    Implicit in the demand integral N_phi=(1-phi)*lambda*integral_0^{v2} f, Section 6; without it, free tier demand could be zero when quality is zero.
  • standard math Standard theorems: IVT, Picard-Lindelof, Lyapunov stability, Leibniz rule, KKT conditions
    Used in proofs in appendices A-F.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Freemium Is All You Need." pith.science (2026). https://pith.science/paper/BV5FNZ7A

@misc{pith2026260800823,
  author       = {Pith},
  title        = {Pith review of: Freemium Is All You Need},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BV5FNZ7A}},
  note         = {Machine review of arXiv:2608.00823}
}
read the original abstract

Free access to a generative AI (GenAI) service requires costly compute, yet can also produce data that improve future service quality. We study a service provider whose users differ in request value and privacy preference, under the constraint that paid requests are kept private, while free requests can be used to improve model quality which otherwise reverts toward a baseline. In particular, we characterize service demand and prove that quality converges to a unique steady state for every stationary service strategy. Next, we analyze optimal pricing and quality policies and show that these can be expressed using two endogenous user value thresholds. For uniform values, we provide a closed-form optimal service strategy and characterize the sufficient conditions for offering free services as dependent on inference cost.

Figures

Figures reproduced from arXiv: 2608.00823 by the authors.

Figure 1
Figure 1. Private requests either pay or exit at v1, while public requests use the free tier below v2 and pay above it. [48] Hal R. Varian. “Chapter 10 Price Discrimination”. In: Handbook of Industrial Orga￾nization. Vol. 1. Elsevier, Jan. 1989, pp. 597–654. doi: 10.1016/S1573- 448X(89) 01013-7. [49] Ander Artola Velasco, Stratis Tsirtsis, Nastaran Okati, and Manuel Gomez-Rodriguez. Is Your LLM Overcharging You? Tokenization,… view at source ↗
Figure 1
Figure 1. Private requests either pay or exit at v1, while public requests use the free tier below v2 and pay above it. A Proofs for Section 4 Proposition 4.6 (Free Tier Necessity). Every mechanism simultaneously satisfying IR, IC, IT, and NS has exactly one positive-quality training-eligible free tier. If more allocations are offered, none charge negative prices, and quality is weakly monotonically increasing in request valu… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

61 extracted references · 4 canonical work pages

  1. [1]

    Optimal Bidding in Multi-Item Multi- Slot Sponsored Search Auctions

    Vibhanshu Abhishek and Kartik Hosanagar. “Optimal Bidding in Multi-Item Multi- Slot Sponsored Search Auctions”. In:Proceedings of the 13th ACM Conference on Electronic Commerce. EC ’12. New York, NY, USA: Association for Computing Ma- chinery, June 2012, p. 2.isbn: 978-1-4503-1415-2.doi:10.1145/2229012.2229015

    arXiv 2012

  2. [2]

    Monetizing Freemium Com- munities: Does Paying for Premium Increase Social Engagement?

    Ravi Bapna, Jui Ramaprasad, and Akhmed Umyarov. “Monetizing Freemium Com- munities: Does Paying for Premium Increase Social Engagement?” In:MIS Quarterly 42.3 (Sept. 2018), pp. 719–735.issn: 0276-7783, 2162-9730.doi:10.25300/MISQ/ 2018/13592

    doi:10.25300/misq/ 2018

  3. [3]

    Social Referral Programs for Freemium Platforms

    Rodrigo Belo and Ting Li. “Social Referral Programs for Freemium Platforms”. In: Management Science68.12 (Dec. 2022), pp. 8933–8962.issn: 0025-1909, 1526-5501. doi:10.1287/mnsc.2022.4301

    arXiv 2022

  4. [4]

    Dirk Bergemann, Alessandro Bonatti, and Alex Smolin.The Economics of Large Language Models: Token Allocation, Fine-Tuning, and Optimal Pricing. Feb. 2025. doi:10.48550/arXiv.2502.07736. arXiv:2502.07736 [econ]

    doi:10.48550/arxiv.2502.07736 2025

  5. [5]

    Deming, Nicola Fuchs-Schündeln, and Jonas Jessen.Mind the Gap: AI Adoption in Europe and the U.S.WorkingPaper.Mar.2026

    Alexander Bick, Adam Blandin, David J. Deming, Nicola Fuchs-Schündeln, and Jonas Jessen.Mind the Gap: AI Adoption in Europe and the U.S.WorkingPaper.Mar.2026. doi:10.3386/w34995. National Bureau of Economic Research:34995

    doi:10.3386/w34995 2026

  6. [6]

    Bernard Bolzano.Rein analytischer Beweis des Lehrsatzes, daß zwischen je zwey Werthen, die ein entgegengesetztes Resultat gewähren, wenigstens eine reelle Wurzel der Gleichung liege.Prague:GottliebHaase,1817.url:https://dml.cz/bitstream/ handle/10338.dmlcz/400016/Bolzano_24-1817-1_1.pdf

  7. [7]

    How Is ChatGPT’s Behavior Chang- ing Over Time?

    Lingjiao Chen, Matei Zaharia, and James Zou. “How Is ChatGPT’s Behavior Chang- ing Over Time?” In:Harvard Data Science Review6.2 (Apr. 2024).issn: 2644-2353, doi:10.1162/99608f92.5317da47

    doi:10.1162/99608f92.5317da47 2024

  8. [8]

    Personalized Pricing in the Presence of Privacy Concerns

    Zhiqi Chen and Mengyu Zhang. “Personalized Pricing in the Presence of Privacy Concerns”. In:Management Science(May 2026).issn: 0025-1909.doi:10 . 1287 / mnsc.2024.07170

    arXiv 2026

Show all 61 references
  1. [9]

    AdsThatStick:Near-OptimalAdOptimizationthroughPsychologicalBehavior Models

    Kailash Gopal Darmasubramanian, Akash Pareek, Arindam Khan, and Arpit Agar- wal.“AdsThatStick:Near-OptimalAdOptimizationthroughPsychologicalBehavior Models”. In:The Fourteenth International Conference on Learning Representations. Oct. 2025.url:https://openreview.net/forum?id=A...

    2025

  2. [10]

    Definite and Semidefinite Quadratic Forms

    Gérard Debreu. “Definite and Semidefinite Quadratic Forms”. In:Econometrica20.2 (1952), pp. 295–300.doi:10.2307/1907852

    1952 doi

  3. [11]

    2025.url:https: //andreyfradkin.com/assets/LLM_Demand_12_12_2025.%5Cgls%7BPDF%7D

    Demirer, Mert, Andrey Fradkin, Tadelis, Nadav, and Sida Tadelis.The Emerging Market for Intelligence: Pricing, Supply, and Demand for LLMs. 2025.url:https: //andreyfradkin.com/assets/LLM_Demand_12_12_2025.%5Cgls%7BPDF%7D

    2025

  4. [12]

    Spillover Effects and Freemium Strategy in the Mobile App Market

    Yiting Deng, Anja Lambrecht, and Yongdong Liu. “Spillover Effects and Freemium Strategy in the Mobile App Market”. In:Management Science69.9 (Sept. 2023), pp. 5018–5041.issn: 0025-1909, 1526-5501.doi:10.1287/mnsc.2022.4619. 15

    2023

  5. [13]

    Mech- anism Design for Large Language Models

    PaulDütting,VahabMirrokni,RenatoPaesLeme,HaifengXu,andSongZuo.“Mech- anism Design for Large Language Models”. In:Proceedings of the ACM Web Confer- ence 2024. WWW ’24. New York, NY, USA: Association for Computing Machinery, May 2024, pp. 144–155.isbn: 979-8-4007-0171-9.doi:10....

    2024

  6. [14]

    DynamicOptimalControl Models in Advertising: Recent Developments

    GustavFeichtinger,RichardF.Hartl,andSureshP.Sethi.“DynamicOptimalControl Models in Advertising: Recent Developments”. In:Management Science40.2 (Feb. 1994), pp. 195–226.issn: 0025-1909, 1526-5501.doi:10.1287/mnsc.40.2.195

    1994 doi

  7. [15]

    Differentiation Under the Integral Sign

    Harley Flanders. “Differentiation Under the Integral Sign”. In:The American Mathe- matical Monthly80.6 (1973), pp. 615–627.doi:10.1080/00029890.1973.11993339

    1973

  8. [16]

    From Competition to Collaboration: Designing Sustainable Mechanisms Between LLMs and Online Forums

    Niv Fono, Yftah Ziser, and Omer Ben-Porat. “From Competition to Collaboration: Designing Sustainable Mechanisms Between LLMs and Online Forums”. In:Proceed- ings of the 25th International Conference on Autonomous Agents and Multiagent Sys- tems. AAMAS ’26. Richland, SC: Intern...

    2026 doi

  9. [17]

    Andrey Fradkin.Demand for LLMs: Descriptive Evidence on Substitution, Market Expansion, and Multihoming. Apr. 2025.doi:10.48550/arXiv.2504.15440. arXiv: 2504.15440 [cs.CY]

    2025 doi arXiv

  10. [18]

    Prediction-Sharing During Training and Inference

    Yotam Gafni, Ronen Gradwohl, and Moshe Tennenholtz. “Prediction-Sharing During Training and Inference”. In:Algorithmic Game Theory. Ed. by Guido Schäfer and Carmine Ventre. Cham: Springer Nature Switzerland, 2024, pp. 425–442.isbn: 978- 3-031-71033-9.doi:10.1007/978-3-031-71033-9_24

    2024 doi

  11. [19]

    Long-Term Data Sharing under Exclusivity Attacks

    Yotam Gafni and Moshe Tennenholtz. “Long-Term Data Sharing under Exclusivity Attacks”. In:Proceedings of the 23rd ACM Conference on Economics and Computa- tion. EC ’22. New York, NY, USA: Association for Computing Machinery, July 2022, pp. 739–759.isbn: 978-1-4503-9150-4.doi:1...

    2022

  12. [20]

    Google AI for Developers

    Google.Gemini API Additional Terms of Service. Google AI for Developers. 2026. url:https://ai.google.dev/gemini-api/terms

    2026

  13. [21]

    Google AI for Developers

    Google.Gemini Developer API Pricing. Google AI for Developers. 2026.url:https: //ai.google.dev/gemini-api/docs/pricing

    2026

  14. [22]

    StabilityinCompetition

    HaroldHotelling.“StabilityinCompetition”.In:The Economic Journal39.153(1929), pp. 41–57.issn: 00130133, 14680297.doi:10.2307/2224214. JSTOR:2224214

    1929 doi

  15. [23]

    Nonparametric Estimation from Incomplete Observations

    Edward L. Kaplan and Paul Meier. “Nonparametric Estimation from Incomplete Observations”. In:Journal of the American Statistical Association53.282 (1958), pp. 457–481.doi:10.1080/01621459.1958.10501452

    1958

  16. [24]

    Strategic Content Creation with GenAI: To Share or Not to Share?

    Gur Keinan and Omer Ben-Porat. “Strategic Content Creation with GenAI: To Share or Not to Share?” In:Proceedings of the ACM Web Conference 2026. WWW ’26. New York, NY, USA: Association for Computing Machinery, Apr. 2026, pp. 169–180.isbn: 979-8-4007-2307-0.doi:10.1145/3774904....

    2026

  17. [25]

    Nonlinear Programming

    Harold W. Kuhn and Albert W. Tucker. “Nonlinear Programming”. In:Proceed- ings of the Second Berkeley Symposium on Mathematical Statistics and Probabil- ity. Ed. by Jerzy Neyman. Berkeley: University of California Press, 1951, pp. 481– 492.url:https : / / projecteuclid . org /...

    1951

  18. [26]

    SomeExtensions of Liapunov’sSecond Method

    Joseph P. LaSalle. “SomeExtensions of Liapunov’sSecond Method”. In:IRE Transac- tions on Circuit Theory7.4 (1960), pp. 520–527.doi:10.1109/TCT.1960.1086720

    1960

  19. [27]

    Fine-Tuning Games: Bargaining and Adaptation for General-Purpose Models

    Benjamin Laufer, Jon Kleinberg, and Hoda Heidari. “Fine-Tuning Games: Bargaining and Adaptation for General-Purpose Models”. In:Proceedings of the ACM Web Con- ference 2024.WWW’24.NewYork,NY,USA:AssociationforComputingMachinery, May 2024, pp. 66–76.isbn: 979-8-4007-0171-9.doi:...

    2024

  20. [28]

    Two-Sided Price Discrimination by Media Platforms

    Song Lin. “Two-Sided Price Discrimination by Media Platforms”. In:Marketing Sci- ence39.2 (Mar. 2020), pp. 317–338.issn: 0732-2399, 1526-548X.doi:10 . 1287 / mksc.2019.1211

    2020

  21. [29]

    Sur l’application de la méthode des approximations succes- sives aux équations différentielles ordinaires du premier ordre

    Ernst Leonard Lindelöf. “Sur l’application de la méthode des approximations succes- sives aux équations différentielles ordinaires du premier ordre”. In:Comptes rendus hebdomadaires des séances de l’Académie des sciences118 (1894), pp. 454–457.url: https://gallica.bnf.fr/ark:/...

  22. [30]

    Pay to (Not) Play: Monetizing Impatience in Mobile Games

    Taylor Lundy, Narun Raman, Hu Fu, and Kevin Leyton-Brown. “Pay to (Not) Play: Monetizing Impatience in Mobile Games”. In:Proceedings of the AAAI Conference on Artificial Intelligence38.9 (Mar. 2024), pp. 9856–9864.issn: 2374-3468.doi:10. 1609/aaai.v38i9.28846

    2024

  23. [31]

    The General Problem of the Stability of Motion

    Aleksandr M. Lyapunov. “The General Problem of the Stability of Motion”. In:Inter- national Journal of Control55.3 (1992). English translation by A. T. Fuller, pp. 531– 773.doi:10.1080/00207179208934253

    1992 doi

  24. [32]

    Strategic Generative AI for Machine Learn- ing in Economic Environments

    Omer Madmon and Moshe Tennenholtz. “Strategic Generative AI for Machine Learn- ing in Economic Environments”. In:ACM AI Lett.1.1 (Mar. 2026), 7:1–7:8.doi: 10.1145/3799994

    2026 doi

  25. [33]

    Optimizing Free-to-Play Multiplayer Games with Premium Subscription

    Yunke Mai and Bin Hu. “Optimizing Free-to-Play Multiplayer Games with Premium Subscription”. In:Management Science69.6 (June 2023), pp. 3437–3456.issn: 0025- 1909, 1526-5501.doi:10.1287/mnsc.2022.4510

    2023

  26. [34]

    Monopoly with Incomplete Information

    Eric Maskin and John Riley. “Monopoly with Incomplete Information”. In:Rand Journal of Economics15.2 (1984), p. 171.issn: 07416261.doi:10.2307/2555674

    1984 doi

  27. [35]

    The Value of Per- sonal Information in Online Markets with Endogenous Privacy

    Rodrigo Montes, Wilfried Sand-Zantman, and Tommaso Valletti. “The Value of Per- sonal Information in Online Markets with Endogenous Privacy”. In:Management Science65.3 (Mar. 2019), pp. 1342–1362.issn: 0025-1909, 1526-5501.doi:10.1287/ mnsc.2017.2989. 17

    2019

  28. [36]

    Incentives in Federated Learning: Equilibria, Dynamics, and Mechanisms for Welfare Maximization

    Aniket Murhekar, Zhuowen Yuan, Bhaskar Ray Chaudhury, Bo Li, and Ruta Mehta. “Incentives in Federated Learning: Equilibria, Dynamics, and Mechanisms for Welfare Maximization”. In:Advances in Neural Information Processing Systems. Vol. 36. Cur- ran Associates, Inc., 2023, pp. 1...

    2023

  29. [37]

    Monopoly and Product Quality

    Michael Mussa and Sherwin Rosen. “Monopoly and Product Quality”. In:Journal of Economic Theory18.2 (Aug. 1978), pp. 301–317.issn: 0022-0531.doi:10.1016/ 0022-0531(78)90085-6

    1978

  30. [38]

    Optimal Auction Design

    Roger B. Myerson. “Optimal Auction Design”. In:Mathematics of Operations Re- search6.1 (1981), pp. 58–73.doi:10.1287/moor.6.1.58

    1981 doi

  31. [39]

    Optimal Advertising Policy under Dynamic Conditions

    Marc Nerlove and Kenneth J. Arrow. “Optimal Advertising Policy under Dynamic Conditions”. In:Economica29.114 (May 1962), p. 129.issn: 00130427.doi:10 . 2307/2551549. JSTOR:10.2307/2551549

    1962 doi

  32. [40]

    Economics of Free under Perpetual Licensing: Implications for the Software Industry

    Marius F. Niculescu and D. J. Wu. “Economics of Free under Perpetual Licensing: Implications for the Software Industry”. In:Information Systems Research25.1 (Mar. 2014), pp. 173–199.issn: 1047-7047, 1526-5536.doi:10.1287/isre.2013.0508

    2014

  33. [41]

    Hamidah Oderinwale and Anna Kazlauskas.The Economics of AI Training Data: A Research Agenda. Apr. 2026.doi:10.48550/arXiv.2510.24990. arXiv:2510.24990 [cs.CY]

    2026 doi arXiv

  34. [42]

    ApproximateMechanismDesignwithout Money

    ArielD.ProcacciaandMosheTennenholtz.“ApproximateMechanismDesignwithout Money”. In:ACM Transactions on Economics and Computation1.4 (Dec. 2013), 18:1–18:26.issn: 2167-8375.doi:10.1145/2542174.2542175

    2013

  35. [43]

    Haoran Sun, Yusen Wu, Peng Wang, Wei Chen, Yukun Cheng, Xiaotie Deng, and Xu Chu.Game Theory Meets Large Language Models: A Systematic Survey with Taxonomy and New Frontiers. Feb. 2025.doi:10.48550/arXiv.2502.09053

    2025 doi arXiv

  36. [44]

    Economics of Content Creation with Generative AI

    Boaz Taitler. “Economics of Content Creation with Generative AI”. In:Proceedings of the AAAI Conference on Artificial Intelligence40.48 (Mar. 2026), pp. 41086–41087. issn: 2374-3468.doi:10.1609/aaai.v40i48.42169

    2026 doi

  37. [45]

    Braess’s Paradox of Generative AI

    Boaz Taitler and Omer Ben-Porat. “Braess’s Paradox of Generative AI”. In:Proceed- ings of the AAAI Conference on Artificial Intelligence39.13 (Apr. 2025), pp. 14139– 14147.issn: 2374-3468.doi:10.1609/aaai.v39i13.33548

    2025 doi

  38. [46]

    May 2025.doi:10.48550/arXiv.2505.12386

    Boaz Taitler, Omer Madmon, Moshe Tennenholtz, and Omer Ben-Porat.Data Shar- ing with a Generative AI Competitor. May 2025.doi:10.48550/arXiv.2505.12386. arXiv:2505.12386 [cs]

    2025 doi arXiv

  39. [47]

    The Theory of Industrial Organization

    Jean Tirole. “The Theory of Industrial Organization”. In:MIT Press Books1 (1988). url:https://ideas.repec.org//b/mtp/titles/0262200716.html

    1988

  40. [48]

    Chapter 10 Price Discrimination

    Hal R. Varian. “Chapter 10 Price Discrimination”. In:Handbook of Industrial Orga- nization. Vol. 1. Elsevier, Jan. 1989, pp. 597–654.doi:10.1016/S1573- 448X(89) 01013-7. 18

    1989 doi

  41. [49]

    Is Your LLM Overcharging You? Tokenization, Transparency, and Incentives

    AnderArtolaVelasco,StratisTsirtsis,NastaranOkati,andManuelGomez-Rodriguez. Is Your LLM Overcharging You? Tokenization, Transparency, and Incentives. Sept. 2025.doi:10.48550/arXiv.2505.21627. arXiv:2505.21627 [cs]

    2025 doi arXiv

  42. [50]

    Ruijie Wang and Aviv Yaish.Order Auctions with Private Position Preferences. Aug. 2026.doi:10.48550/ARXIV.2608.00786

    2026 doi arXiv

  43. [51]

    The Non-Local Existence Problem of Ordinary Differential Equa- tions

    Aurel Wintner. “The Non-Local Existence Problem of Ordinary Differential Equa- tions”. In:American Journal of Mathematics67.2 (1945), pp. 277–284.doi:10 . 2307/2371729

    1945

  44. [52]

    Heterogeneous Data Game: Characterizing the Model Competition Across Multiple Data Sources

    Renzhe Xu, Kang Wang, and Bo Li. “Heterogeneous Data Game: Characterizing the Model Competition Across Multiple Data Sources”. In:Forty-Second International Conference on Machine Learning. June 2025, p. 24.url:https : / / openreview . net/forum?id=o36quGYoQX

    2025

  45. [53]

    Weijie Zhong.Token Is All You Price. Oct. 2025.doi:10.48550/arXiv.2510.09859. arXiv:2510.09859 [econ]. 19 v v1 Privateρ= 1 No Use Paid v v2 Publicρ= 0 Free Paid Figure 1: Private requests either pay or exit atv1, while public requests use the free tier belowv 2 and pay above i...

    2025 doi arXiv

  46. [54]

    (16), giving Proposition 10.5

    This leads us to choose the negative root in Eq. (16), giving Proposition 10.5. Having now chosen the negative root, we are guaranteedx∗ 1 >0. Proposition 10.6.The unconstrained maximizers Proposition 10.5 are optimal if both x∗ 1 < x∗ 2 andx ∗ 2 ≤1hold. These constraints resp...

  47. [55]

    If Eq.(8)and Eq.(9)hold thenx ∗ 1, x∗ 2 follow Proposition 10.5

  48. [56]

    If Eq.(8)is violated while Eq.(9)holds, thenx ∗ 1 =x ∗ 2 = 1+ √ 1+3 Cφ KM 3 ,andγ= 0

  49. [57]

    If Eq.(9)is violated and Cφ KM < 3φ−1 4 , thenx ∗ 2 = 1,and Proposition 10.5 holds

  50. [58]

    If Eq.(9)is violated and Cφ KM ≥ 3φ−1 4 , thenx ∗ 1 =x ∗ 2 = 1, andγ= 0. Proof.1. If both Eq. (8) and Eq. (9) hold, then the optimalx ∗ 1, x∗ 2 are not constrained and are given by Proposition 10.5, Proposition 10.5. Note that this is possible, as (3φ−2)< −1+2φ (2−φ)2, holds f...

  51. [59]

    (8) is violated while Eq

    If Eq. (8) is violated while Eq. (9) holds, then we have that Cφ KM ≥ −1+2φ (2−φ)2. This will result in the constraintx∗ 1 ≤x ∗ 2 being violated. This means that we must set x∗ 1 =x ∗ 2,and both will satisfyx ∗ 1, x∗ 2 ≤1also resulting inγ= 0. In this case the objective Propos...

  52. [60]

    (9) is violated, that isCφ KM <(3φ−2), means thatx ∗ 2 ≥1, so we must setx ∗ 2 = 1

    Similarly, in case Eq. (9) is violated, that isCφ KM <(3φ−2), means thatx ∗ 2 ≥1, so we must setx ∗ 2 = 1. We then need to check weatherx∗ 1 <1, i.e., 2− √ 1−12 Cφ KM (1−φ) 3φ <1. This in turn is equivalent to q 1−12 Cφ KM (1−φ)>3φ−2. This holds if12 Cφ KM (1− φ)<1−(3φ−2) 2 =−...

  53. [61]

    exclusivity

    Otherwise, if both Eq. (9) and Eq. (10) are violated thenx∗ 1 =x ∗ 2 = 1, again resulting inγ= 0. The last scenario, namelyx ∗ 1 =x ∗ 2 = 1, withγ= 0also happens when both Eq. (8) and Eq. (9) are violated. F Additional Related Work Privacy and Platform Design.Previous work exp...

Pith tools

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