Pith. sign in

REVIEW 5 major objections 5 minor 43 references

BiMarker: Enhancing Text Watermark Detection for Large Language Models with Bipolar Watermarks

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

Pith's one-line read Splitting watermarked text into two oppositely-biased poles and comparing green-token counts between them gives a higher z-score lower bound than KGW at no extra false positives, the paper claims.

desk verdict BiMarker's differential-counting idea and its gamma=0.5 experiments are genuinely worth a look, but the paper's central theorem is unproven as written because the proof and Algorithm 1 use different green-list conventions. read the letter →

arxiv 2501.12174 v6 pith:4K5K2PAZ submitted 2025-01-21 cs.LG

classification cs.LG
keywords LLMtextwatermarkingbipolarwatermarkdifferentialdetectionstatisticgreen-listlogitboostingfalsepositiverateKGWentropy-baseddetectorswithoutprompt
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

BiMarker attacks a weakness in the standard KGW watermark: its detector compares a text's green-token count against a fixed expected value $\gamma T$, but real human-written text fluctuates around that value and tends to track the watermarked distribution, which blunts detection. The paper's fix is to mark each generation position as a positive or negative pole: in positive positions the green-list logits are boosted by $\delta$ exactly as in KGW, while in negative positions the red-list logits are boosted instead. Detection then uses the difference between the two poles' green counts, so the unknown baseline of human text cancels within the text itself rather than being estimated from outside. The paper proves that this differential statistic has a higher theoretical lower bound than KGW's (Theorem 3.1) without raising the false-positive rate (Theorem 3.2), and reports higher true-positive rates at matched false-positive rates, with the largest gains at low watermark strength, while keeping text perplexity essentially unchanged because the same $\gamma$ and $\delta$ perturb the same share of the vocabulary. If the claims hold, the payoff is a free detection gain: better separation of AI text from human text with no added inference cost, no prompt access, and direct compatibility with entropy-based detectors such as SWEET and EWD.

What carries the argument

The carrying object is the differential z-statistic over a bipolar partition of the text, along with the polarity-embedding rule in Algorithm 1. At each position the embedding assigns a polarity pseudo-randomly with probability $\rho$ of being positive; in positive positions the green list (a random $\gamma|V|$ subset of the vocabulary) has its logits raised by $\delta$, and in negative positions the complementary red list is raised by $\delta$, so the same $\gamma$ share of the vocabulary is perturbed throughout and the token-level distribution change matches KGW. During detection, green tokens are counted within each pole and the statistic subtracts the null expectation $\gamma T_p - (1-\gamma) T_n$; at the ratio $T_p/T_n = (1-\gamma)/\gamma$ the statistic reduces to $(|s|^p_G - |s|^n_G)/\sqrt{T\gamma(1-\gamma)}$. What does the work is the cancellation of the human-text baseline within the same piece of text, replacing KGW's fixed external reference $\gamma T$, plus two supporting results: Theorem 3.1, which lower-bounds the watermarked z-score via the $\alpha = \exp(\delta)$ boosting bound of Lemma A.1, and Theorem 3.2, which bounds the differential false-positive rate by the unipolar rate using Gaussian tail comparison. The same differential statistic is then applied, unchanged in structure, to the entropy-gated token set of SWEET and to the entropy-weighted terms of EWD.

What would settle it

Set $\gamma=0.25$ (or any value away from 0.5) with $T_p/T_n = (1-\gamma)/\gamma$ and check two things. First, simulate the null by applying Algorithm 1's list partitions to thousands of genuinely non-watermarked sequences and estimate the empirical false-positive rate of the Eq. (1) statistic at a KGW-matched threshold; if the statistic is visibly off-center or the FPR exceeds KGW's, Theorem 3.2 fails under the algorithm's own convention. Second, recompute the Theorem 3.1 bound difference, Eqs. (16)-(17), substituting the algorithm's actual negative-pole green list of size $(1-\gamma)|V|$ where the proof uses a list of size $\gamma|V|$, and check whether $B(|z|_d) - B(|z|_k) \geq 0$ still holds over the full range of spike-entropy values; if it turns negative for some values, the claimed strictly higher lower bound is not established.

Watch

Extended reading notes

Core claim

BiMarker is a watermarking scheme in which the generating rule itself supplies the baseline. The text is divided into positive and negative poles; the positive pole upweights a random $\gamma|V|$ green list by $\delta$, and the negative pole upweights the complementary red list by the same $\delta$. The detector counts green tokens separately in each pole and forms the differential statistic $z = (|s|^p_G - |s|^n_G - \gamma T_p + (1-\gamma) T_n)/\sqrt{T\gamma(1-\gamma)}$, which, when the pole sizes satisfy $T_p/T_n = (1-\gamma)/\gamma$, collapses to the bare difference of the two green counts divided by $\sqrt{T\gamma(1-\gamma)}$; the null expectation is then zero for any human text regardless of its overall green tendency, and it is this cancellation that the paper identifies as the source of KGW's vulnerability to misestimating non-watermarked text. The paper's central claim is that the differential detector attains a higher theoretical lower bound on the z-score of watermarked text than KGW's single-pole statistic (Theorem 3.1, proved from the green-list probability bound of Lemma A.1), that it does not increase the false-positive rate over non-watermarked text (Theorem 3.2, by Gaussian tail comparison), and that both theorems survive when the differential scheme is layered onto the entropy-gated detector SWEET and the entropy-weighted detector EWD. Because embedding still boosts exactly a $\gamma$ share of the vocabulary by the same $\delta$, the induced change in the generated distribution matches KGW's, so the method is argued to preserve text quality while sharpening detectability.

Load-bearing premise

The load-bearing premise is the detector's null model for human text, namely that the two poles' green-token counts are independent and centered at the differential expectation $\gamma T_p - (1-\gamma)T_n$, together with one consistent definition of the negative pole's 'green list' across the algorithm, the statistic, and the proof; the paper's algorithm and its proof use different definitions of that list whenever $\gamma \neq 0.5$, so the false-positive guarantee is clean only where the two conventions coincide.

Editorial extensions

If this is right

  • At a matched false-positive budget, watermarked text is detected more often: at 1% FPR with $\gamma=0.5$ and multinomial sampling, TPR rises from 0.436 to 0.498 at $\delta=0.5$ and from 0.978 to 0.986 at $\delta=1.5$.
  • The largest gains sit exactly where KGW is weakest, namely low watermark strength and strict (1% or zero) false-positive targets, so a practitioner can use a smaller $\delta$ (less quality cost) for the same reliability.
  • Detection needs no prompt and no auxiliary language model: the within-text pole difference replaces the external $\gamma T$ baseline, so the method applies when the prompt is unavailable and at the same computational cost as KGW.
  • Because embedding perturbs the same $\gamma$ share of the vocabulary with the same $\delta$, the text-quality impact equals KGW's, so the detection gain is argued to come without a generation-quality penalty.
  • The differential statistic composes with entropy-based optimizations: on code tasks, EWD with BiMarker reaches a best F1 of 0.96 on HumanEval versus 0.869 for EWD alone, with the paper's theorems stated to carry over to SWEET and EWD.

Reading between the lines

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

  • Editorial extension: the cancellation trick should transfer to any watermark that biases a fixed vocabulary split, such as multi-color, semantic, or multi-bit green lists, because what BiMarker replaces is only the external null mean, not the KGW mechanism itself; this is testable by swapping the embedding while keeping the differential detector.
  • Editorial caveat: the proof of Theorem 3.1 bounds the negative pole's green count as if its green list had size $\gamma|V|$, whereas Algorithm 1 defines the negative-pole green list as the complementary set of size $(1-\gamma)|V|$; the two conventions coincide only at $\gamma=0.5$, so a re-derivation of the bounds in the algorithm's own convention is the natural next step before relying on the exa
  • Editorial extension: the paper's motivation study reports a Pearson correlation of 0.7 between human and watermarked green-token counts; a stress test across other domains (multilingual text, dialogue, code) would show whether the differential gain shrinks when that correlation is weak.
  • Editorial extension: the hard-coded position-based polarity used with SWEET and EWD anchors the signal to absolute token positions, so insertions, deletions, or reordering that desynchronize the polarity clock could degrade the differential signal more than they degrade KGW's single-pole signal; the paper's limitations section concedes related fragility from adjacent-token dependencies in code.
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.

Referee Report

5 major / 5 minor

Summary. The paper proposes BiMarker, a KGW-style watermarking method that alternates between two poles: in the positive pole the usual gamma-fraction green list is boosted, while in the negative pole the complementary list (size 1-gamma) is treated as green and the other list is boosted. Detection uses a differential z-score that compares green-token counts in the two poles. The paper claims that this differential statistic has a higher theoretical lower bound than KGW's z-score without increasing the false positive rate (Theorems 3.1 and 3.2), and it reports experiments on news-like text and code generation showing improved TPR at fixed FPR, especially at gamma=0.5, as well as compatibility with SWEET and EWD and preserved text quality (PPL, pass@1).

Significance. If the theoretical claims were correct, BiMarker would be a practically attractive, zero-overhead enhancement to KGW. The paper should be credited for clean empirical comparisons at gamma=0.5, quality checks (PPL and pass@1), robustness evaluation via back-translation, and for demonstrating that the bipolar construction can be combined with SWEET and EWD. However, the theoretical core is not established: the proof of Theorem 3.1 is inconsistent with Algorithm 1, the final algebra in the proof is incorrect for gamma != 0.5, the entropy-maximization step uses the wrong direction of convexity, and the proof of Theorem 3.2 contains formal errors. Since the theoretical analysis is presented as a main contribution and is used to explain the experimental gains, these issues are load-bearing.

major comments (5)
  1. [Appendix C.1, Eq. (14)] The proof of Theorem 3.1 uses a negative-pole green-count bound that does not correspond to Algorithm 1. In Algorithm 1, when the polarity is negative, the green list is list2 of size (1-gamma)|V| and the logits of list1 (size gamma|V|) are boosted. Equation (14) instead bounds the expected negative-pole green count by gamma*Tn/(1+(alpha-1)*gamma), which is the value obtained when a green list of size gamma|V| is boosted. Under the Algorithm 1 convention, the negative-pole green probability in the uniform case is (1-gamma)*Tn/(1+(alpha-1)*gamma), which is larger than the proof's bound whenever gamma<0.5, and for non-uniform token distributions the discrepancy can be much larger. Therefore the lower bound in Eq. (15) is not a valid bound for the method defined in Algorithm 1.
  2. [Appendix C.1, Eqs. (16)-(17)] Even under the proof's own convention, the final step '>= ... = 0' in Eq. (17) is algebraically valid only when gamma=0.5. The expression equals Tn*[1 - gamma*(alpha+1)/(1+(alpha-1)*gamma)] / sqrt(T*gamma*(1-gamma)), which is negative for gamma>0.5. Since Algorithm 1 and Theorem 3.1 allow any gamma in (0,1), the claim that BiMarker's lower bound 'consistently exceeds' KGW's lower bound is false as stated; at a minimum a condition such as gamma <= 0.5 would need to be stated and justified.
  3. [Appendix C.1, paragraph before Eq. (14)] The proof treats the uniform token distribution as the case that maximizes the negative-pole green count and uses it as an upper bound. This is backwards: for a fixed green-list size g, the green probability under red-list boosting is P_G/[1+(alpha-1)*(1-P_G)], which is convex in P_G for alpha>1. Jensen's inequality then implies that the uniform case gives a lower bound on the expected green probability, not an upper bound. Consequently, subtracting gamma*Tn/(1+(alpha-1)*gamma) in Eq. (14) is not conservative, and the argument cannot establish a lower bound on the differential z-score.
  4. [Appendix C.2, Eqs. (20)-(22)] The proof of Theorem 3.2 is not rigorous as written. Equation (21) is missing the density f(x) in the integrand and uses mu_Tn in the numerator where mu_Tp is required. Equation (22) then upper-bounds the integrand by the constant 1-Phi(t/sigma_T) and appears to integrate it over (-infinity, infinity), which diverges; the claim that 'x - gamma*Tn > 0' is not true on the entire integration range. The asserted relationship F_KGW >= F_Diff is therefore not established by the given argument.
  5. [Section 3.3, Table 1, Figure 3] Under the iid model assumed in Theorem 3.1 with gamma=0.5 and Tp=Tn, the BiMarker differential statistic is exactly equivalent to the KGW statistic: the negative-pole green count can be written as T/2 minus a Binomial(T/2,p_p) variable, so G_p - G_n has the same distribution as KGW's G - T/2. Thus no strict improvement in the z-score distribution can arise in this model. The empirical gains in Table 1 and Figure 3 must therefore come from a mechanism not captured by Theorem 3.1 (for example, hard-coded polarity, null-mean misspecification, or non-iid text structure), and the paper should state and analyze that mechanism explicitly.
minor comments (5)
  1. [Section 3.1 and 3.2] The heading 'Movitation' should be 'Motivation', and the sentence 'Notably, our method consistently increases logits across gamma proportion of the vocabulary' is duplicated in Section 3.2; one copy should be removed.
  2. [Section 4] The main high-entropy experiments do not specify whether polarity assignment uses the pseudo-random rule of Algorithm 1 or the position-based hard-coded rule described later. This matters because random polarity makes Tp and Tn random, which changes the null variance of the statistic in Eq. (1).
  3. [Table 2] The column header 'B EST' is unclear and is presumably intended to be 'BEST'; please clarify.
  4. [Appendix C.2] The symbol t is defined in Eq. (20) as z_threshold * sqrt(T*gamma*(1-gamma)) and then redefined in the EWD discussion; using distinct symbols would avoid confusion.
  5. [References] Langley (2000) appears in the bibliography but is not cited in the text, and Holtzman et al. (2019) is cited only in Appendix D.1; please clean up the citation list.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: BiMarker's differential statistic is anchored to the external KGW baseline and derived from stated green-list definitions, with no fitted parameter renamed as a prediction.

full rationale

Walking the derivation chain, I find no circular step that reduces a prediction to an input. BiMarker's differential z-statistic (Eq. 1) is defined directly from the null expected difference Tp·γ − Tn·(1−γ), which follows from the green/red list sizes in Algorithm 1; it is not fitted to the data that the paper later predicts. The central theoretical claim, Theorem 3.1, is an algebraic comparison of lower bounds obtained from KGW's Lemma A.1, an external baseline result, and the proof does not rely on a self-citation chain. The optimal polarity ratio ρ = (1−γ)/γ is derived from the stated variance-minimizing condition Tp/Tn = (1−γ)/γ, then validated experimentally across ρ values rather than fit as a free parameter. The paper's experimental comparisons against KGW, SWEET, and EWD are external benchmarks, and the method requires no fitted constants to operate. There is an apparent rigor concern in Appendix C.1: the proof uses an unproved equality S* = S*_p and appears to treat the negative pole's green list as having size γ|V|, while Algorithm 1 assigns negative-pole green tokens to the (1−γ)|V| list; however, this is an internal correctness/proof-consistency issue, not a circular reduction of the theorem to its assumptions. The limitations section candidly acknowledges hard-coded polarity risks and token-dependence assumptions, which further supports that the paper is not concealing a circular derivation. Accordingly, the appropriate circularity finding is none, with score 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The scorecard relies on standard KGW machinery (Gaussian null model, spike entropy bounds), two domain assumptions about independent poles and correlated human/watermarked green counts, and one ad hoc assumption in the proof of Theorem 3.1 (S*=S*p). The main methodological parameters (gamma, delta, rho, polarity cycle lengths, entropy threshold) are hyperparameters, not fitted to data. No invented entities.

free parameters (5)
  • gamma (green-list fraction) = 0.25, 0.5 in main experiments
    Inherited from KGW; controls list sizes and detection statistics. Chosen by hand from prior practice, not fitted.
  • delta (watermark strength) = 0.25 to 2.5 in main experiments
    Inherited from KGW; controls bias magnitude. Swept across values; not fitted.
  • rho (positive polarity probability) = 0.5 (gamma=0.5), 0.75 (gamma=0.25), swept 0.1-0.9 in Figure 6
    Algorithm 1 takes rho as input; the paper derives the optimal rho=1-gamma and validates it empirically, so it is a tunable hyperparameter rather than a fitted constant.
  • polarity cycle length for code tasks = 20 positive/20 negative tokens for KGW and EWD; 15/15 for SWEET
    Hand-chosen in Section 4 for low-entropy code generation; the authors note integrity risks from this choice.
  • entropy threshold tau (SWEET) = 0.695
    Adopted from prior work (Lee et al., Lu et al.), not derived in this paper.
assumptions (5)
  • domain assumption Green-token counts in non-watermarked text follow a Gaussian distribution under the null hypothesis.
    Adopted from KGW and used in Theorem 3.2's FPR analysis (Appendix C.2).
  • domain assumption Positive and negative pole token counts are independent and Gaussian with variances sigma_Tp^2 and sigma_Tn^2.
    Stated in Section 3.2 and used in Theorem 3.2.
  • ad hoc to paper Average spike entropy of the whole sequence equals that of the positive pole (S*=S*p).
    Introduced in the proof of Theorem 3.1 (Appendix C.1, after Eq. 16) to simplify the lower-bound comparison; not justified in the paper.
  • standard math Spike entropy is at most 1.
    Used in Eq. (17) and the EWD discussion to bound S* from above by 1.
  • domain assumption Human-written and watermarked green-token counts follow a similar trend across contexts.
    Core motivation in Section 3.1, empirically supported by a Pearson correlation of 0.7 computed on a selected subset.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BiMarker: Enhancing Text Watermark Detection for Large Language Models with Bipolar Watermarks." pith.science (2026). https://pith.science/paper/4K5K2PAZ

@misc{pith2026250112174,
  author       = {Pith},
  title        = {Pith review of: BiMarker: Enhancing Text Watermark Detection for Large Language Models with Bipolar Watermarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4K5K2PAZ}},
  note         = {Machine review of arXiv:2501.12174}
}
read the original abstract

The rapid growth of Large Language Models (LLMs) raises concerns about distinguishing AI-generated text from human content. Existing watermarking techniques, like \kgw, struggle with low watermark strength and stringent false-positive requirements. Our analysis reveals that current methods rely on coarse estimates of non-watermarked text, limiting watermark detectability. To address this, we propose Bipolar Watermark (\tool), which splits generated text into positive and negative poles, enhancing detection without requiring additional computational resources or knowledge of the prompt. Theoretical analysis and experimental results demonstrate \tool's effectiveness and compatibility with existing optimization techniques, providing a new optimization dimension for watermarking in LLM-generated content.

Figures

Figures reproduced from arXiv: 2501.12174 by the authors.

Figure 1
Figure 1. Illustration of the key idea of BiMarker: Adjust token bias by polarity during generation and detect by the green token difference between polarities. Watermarking techniques offer a solution by enabling LLMs to embed unique, imperceptible identifiers (watermarks) within generated content, distinguishing it from human￾written text (Liu et al., 2024b; Pan et al., 2024b). A promi￾nent method, KGW1 , achieves high dete… view at source ↗
Figure 2
Figure 2. Empirical Study of the Relationship Between the Number of Green Tokens in Human-Written Texts and the Number of Green Tokens in Watermarked Texts. 3.1. Movitation Before introducing our method, we first discuss the dis￾tribution of human-written text. It is undeniable that the distribution of green tokens in human-written text should be γT, but in more fine-grained scenarios, this may not always be accurate. Further… view at source ↗
Figure 3
Figure 3. The relationship between watermark strength and accuracy under zero false positives (0/500). The left figure illustrates results with multinomial sampling, while the right figure depicts results using beam search with a beam size of 8 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: The impact of ρ values on detection results. We set δ = 1.5 and used multinomial sampling. tokens [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: The Average Z-Score as a Function of Token Length (T) of the Generated Text with multinomial sampling. 200 · ρ tokens were assigned to the positive polarity, while the remaining tokens were assigned to the negative polar￾ity. The experimental results, shown in [PITH_F…
Figure 7
Figure 7. Figure 7: Detection Impact from Inaccurate Non-Watermarked Text Estimations with Watermarked Text (where γ is 0.5 and δ is 1.0) Generated by OPT-1.3, assuming classification as watermarked text occurs when the z-value exceeds 4.0. B.2. Applying BiMarker to EWD EWD is another wat…
Figure 8
Figure 8. Figure 8: PPL comparison under different watermark strengths (δ ∈ [0.5, 0.75, 1.0, 1.5, 2, 2.5]), where the x-axis denotes BiMarker’s PPL and the y-axis shows the ratio of KGW’s PPL to BiMarker’s. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 17 canonical work pages

  1. [1]

    Program synthesis with large language models

    Austin, J., Odena, A., Nye, M., Bosma, M., Michalewski, H., Dohan, D., Jiang, E., Cai, C., Terry, M., Le, Q., et al. Program synthesis with large language models. arXiv preprint arXiv:2108.07732, 2021

  2. [2]

    M., Gebru, T., McMillan-Major, A., and Shmitchell, S

    Bender, E. M., Gebru, T., McMillan-Major, A., and Shmitchell, S. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, New York, NY, USA, 2021. Association for Computing Machinery

  3. [3]

    S., Abercrombie, G., Spruit, S., Hovy, D., Dinan, E., Boureau, Y.-L., and Rieser, V

    Bergman, A. S., Abercrombie, G., Spruit, S., Hovy, D., Dinan, E., Boureau, Y.-L., and Rieser, V. Guiding the release of safer E 2 E conversational AI through value sensitive design. In Proceedings of the 23rd Annual Meeting of the Special Interest Group on Discourse and Dialogue, September 2022

  4. [4]

    Bad characters: Imperceptible nlp attacks

    Boucher, N., Shumailov, I., Anderson, R., and Papernot, N. Bad characters: Imperceptible nlp attacks. In 2022 IEEE Symposium on Security and Privacy (SP), pp.\ 1987--2004. IEEE, 2022

  5. [5]

    T., Low, S., Maxemchuk, N

    Brassil, J. T., Low, S., Maxemchuk, N. F., and O'Gorman, L. Electronic marking and identification techniques to discourage document copying. IEEE Journal on Selected Areas in Communications, 13 0 (8): 0 1495--1504, 1995

  6. [6]

    T., Li, Y., Lundberg, S., et al

    Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y. T., Li, Y., Lundberg, S., et al. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712, 2023

  7. [7]

    Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. D. O., Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G., et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021

  8. [9]

    Undetectable watermarks for language models

    Christ, M., Gunn, S., and Zamir, O. Undetectable watermarks for language models. arXiv preprint arXiv:2306.09194, 2023 b

Show all 43 references
  1. [10]

    Machine generated text: A comprehensive survey of threat models and detection methods

    Crothers, E., Japkowicz, N., and Viktor, H. Machine generated text: A comprehensive survey of threat models and detection methods. In Proceedings of arXiv, arXiv, 2022. arXiv. doi:10.48550/arXiv.2210.07321. URL http://arxiv.org/abs/2210.07321

  2. [11]

    WordNet: An electronic lexical database

    Fellbaum, C. WordNet: An electronic lexical database. MIT Press, 1998

  3. [12]

    Three bricks to consolidate watermarks for large language models

    Fernandez, P., Chain, A., Tit, K., Chappelier, V., and Furon, T. Three bricks to consolidate watermarks for large language models. In 2023 IEEE International Workshop on Information Forensics and Security (WIFS), pp.\ 1--6. IEEE, 2023

  4. [13]

    and Adomaitis, L

    Grinbaum, A. and Adomaitis, L. The ethical need for watermarks in machine-generated language. In Proceedings of arXiv, arXiv, 2022. arXiv. doi:10.48550/arXiv.2209.03118. URL http://arxiv.org/abs/2209.03118

  5. [14]

    The curious case of neural text degeneration

    Holtzman, A., Buys, J., Du, L., Forbes, M., and Choi, Y. The curious case of neural text degeneration. arXiv preprint arXiv:1904.09751, 2019

  6. [15]

    Jawahar, G., Abdul-Mageed, M., and Lakshmanan, L. V. Automatic detection of machine generated text: A critical survey. arXiv preprint arXiv:2011.01314, 2020

  7. [16]

    A watermark for large language models

    Kirchenbauer, J., Geiping, J., Wen, Y., Katz, J., Miers, I., and Goldstein, T. A watermark for large language models. In International Conference on Machine Learning, 2023 a . URL https://api.semanticscholar.org/CorpusID:256194179

  8. [17]

    On the reliability of watermarks for large language models

    Kirchenbauer, J., Geiping, J., Wen, Y., Shu, M., Saifullah, K., Kong, K., Fernando, K., Saha, A., Goldblum, M., and Goldstein, T. On the reliability of watermarks for large language models. arXiv preprint arXiv:2306.04634, 2023 b

  9. [18]

    Robust distortion-free watermarks for language models

    Kuditipudi, R., Thickstun, J., Hashimoto, T., and Liang, P. Robust distortion-free watermarks for language models. arXiv preprint arXiv:2307.15593, 2023

  10. [19]

    Crafting papers on machine learning

    Langley, P. Crafting papers on machine learning. In Langley, P. (ed.), Proceedings of the 17th International Conference on Machine Learning (ICML 2000), pp.\ 1207--1216, Stanford, CA, 2000. Morgan Kaufmann

  11. [20]

    Who wrote this code? watermarking for code generation

    Lee, T., Hong, S., Ahn, J., Hong, I., Lee, H., Yun, S., Shin, J., and Kim, G. Who wrote this code? watermarking for code generation. Association for Computational Linguistics, 2024. URL https://aclanthology.org/2024.acl-long.268

  12. [21]

    Starcoder: May the source be with you! arXiv preprint arXiv:2305.06161, 2023

    Li, R., Ben Allal, L., Zi, Y., Muennighoff, N., Kocetkov, D., Mou, C., Marone, M., Akiki, C., Li, J., Chim, J., et al. Starcoder: May the source be with you! arXiv preprint arXiv:2305.06161, 2023

  13. [22]

    A semantic invariant robust watermark for large language models

    Liu, A., Pan, L., Hu, X., Meng, S., and Wen, L. A semantic invariant robust watermark for large language models. In The Twelfth International Conference on Learning Representations, 2024 a

  14. [23]

    A survey of text watermarking in the era of large language models

    Liu, A., Pan, L., Lu, Y., Li, J., Hu, X., Zhang, X., Wen, L., King, I., Xiong, H., and Yu, P. A survey of text watermarking in the era of large language models. ACM Computing Surveys, 57 0 (2): 0 1--36, 2024 b

  15. [24]

    An entropy-based text watermarking detection method

    Lu, Y., Liu, A., Yu, D., Li, J., and King, I. An entropy-based text watermarking detection method. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, 2024

  16. [25]

    Munyer, T. J. E. and Zhong, X. Deeptextmark: Deep learning based text watermarking for detection of large language model generated text. ArXiv, abs/2305.05773, 2023. URL https://api.semanticscholar.org/CorpusID:258588289

  17. [26]

    Chatgpt: Optimizing language models for dialogue

    OpenAI. Chatgpt: Optimizing language models for dialogue. https://openai.com/blog/chatgpt/, November 2022. Accessed: 2024-11-25

  18. [27]

    Pan, L., Liu, A., He, Z., Gao, Z., Zhao, X., Lu, Y., Zhou, B., Liu, S., Hu, X., Wen, L., King, I., and Yu, P. S. M ark LLM : An open-source toolkit for LLM watermarking. In Hernandez Farias, D. I., Hope, T., and Li, M. (eds.), Proceedings of the 2024 Conference on Empirical Me...

  19. [28]

    Pan, L., Liu, A., Lu, Y., Gao, Z., Di, Y., Huang, S., Wen, L., King, I., and Yu, P. S. Waterseeker: Pioneering efficient detection of watermarked segments in large documents. arXiv preprint arXiv:2409.05112, 2024 b

  20. [29]

    Y., Wong, K., and Chee, K

    Por, L. Y., Wong, K., and Chee, K. O. Unispach: A text-based data hiding method using unicode space characters. Journal of Systems and Software, 85 0 (5): 0 1075--1082, 2012. doi:10.1016/j.jss.2011.12.023

  21. [30]

    W., Xu, T., Brockman, G., McLeavey, C., and Sutskever, I

    Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., and Sutskever, I. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pp.\ 28492--28518. PMLR, 2023

  22. [31]

    Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., and Liu, P. J. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21 0 (1), January 2020. ISSN 1532-4435

  23. [32]

    A robust semantics-based watermark for large language model against paraphrasing

    Ren, J., Xu, H., Liu, Y., Cui, Y., Wang, S., Yin, D., and Tang, J. A robust semantics-based watermark for large language model against paraphrasing. In Findings of the Association for Computational Linguistics: NAACL 2024. Association for Computational Linguistics, 2024. doi:1...

  24. [33]

    Embarrassingly simple text watermarks

    Sato, R., Takezawa, Y., Bao, H., Niwa, K., and Yamada, M. Embarrassingly simple text watermarks. arXiv preprint, 2023

  25. [34]

    Topkara, U., Topkara, M., and Atallah, M. J. The hiding virtues of ambiguity: quantifiably resilient watermarking of natural language text through synonym substitutions. In Proceedings of the 8th Workshop on Multimedia and Security, pp.\ 164--174, 2006

  26. [35]

    Llama: Open and efficient foundation language models

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozi \`e re, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., and Lample, G. Llama: Open and efficient foundation language models. ArXiv, abs/2302.13971, 2023. URL https...

  27. [36]

    Waterbench: Towards holistic evaluation of watermarks for large language models

    Tu, S., Sun, Y., Bai, Y., Yu, J., Hou, L., and Li, J. Waterbench: Towards holistic evaluation of watermarks for large language models. arXiv preprint arXiv:2311.07138, 2023

  28. [37]

    Towards codable text watermarking for large language models

    Wang, L., Yang, W., Chen, D., Zhou, H., Lin, Y., Meng, F., Zhou, J., and Sun, X. Towards codable text watermarking for large language models. arXiv preprint arXiv:2307.15992, 2023

  29. [38]

    Fairness feedback loops: training on synthetic data amplifies bias

    Wyllie, S., Shumailov, I., and Papernot, N. Fairness feedback loops: training on synthetic data amplifies bias. In The 2024 ACM Conference on Fairness, Accountability, and Transparency, pp.\ 2113--2147, 2024

  30. [39]

    Tracing text provenance via context-aware lexical substitution

    Yang, X., Zhang, J., Chen, K., Zhang, W., Ma, Z., Wang, F., and Yu, N. Tracing text provenance via context-aware lexical substitution. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp.\ 11613--11621, 2022

  31. [40]

    Watermarking text generated by black-box language models

    Yang, X., Chen, K., Zhang, W., Liu, C., Qi, Y., Zhang, J., Fang, H., and Yu, N. Watermarking text generated by black-box language models. arXiv preprint, 2023

  32. [41]

    Advancing beyond identification: Multi-bit watermark for large language models

    Yoo, K., Ahn, W., and Kwak, N. Advancing beyond identification: Multi-bit watermark for large language models. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Pape...

  33. [42]

    V., et al

    Zhang, S., Roller, S., Goyal, N., Artetxe, M., Chen, M., Chen, S., Dewan, C., Diab, M., Li, X., Lin, X. V., et al. Opt: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068, 2022

  34. [43]

    Provable robust watermarking for ai-generated text

    Zhao, X., Ananth, P., Li, L., and Wang, Y.-X. Provable robust watermarking for ai-generated text. arXiv preprint arXiv:2306.17439, 2023

  35. [44]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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