Intrinsic Dimension Estimation for Robust Detection of AI-Generated Texts
Open accessAbstract
Rapidly increasing quality of AI-generated content makes it difficult to distinguish between human and AI-generated texts, which may lead to undesirable consequences for society. Therefore, it becomes increasingly important to study the properties of human texts that are invariant over different text domains and varying proficiency of human writers, can be easily calculated for any language, and can robustly separate natural and AI-generated texts regardless of the generation model and sampling method. In this work, we propose such an invariant for human-written texts, namely the intrinsic dimensionality of the manifold underlying the set of embeddings for a given text sample. We show that the average intrinsic dimensionality of fluent texts in a natural language is hovering around the value $9$ for several alphabet-based languages and around $7$ for Chinese, while the average intrinsic dimensionality of AI-generated texts for each language is $\approx 1.5$ lower, with a clear statistical separation between human-generated and AI-generated distributions. This property allows us to build a score-based artificial text detector. The proposed detector's accuracy is stable over text domains, generator models, and human writer proficiency levels, outperforming SOTA detectors in model-agnostic and cross-domain scenarios by a significant margin.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Tulchinskii et al. (2023) approach is robust across cross-domain challenges, model shifts, and adversarial attacks.
“It is imperative to acknowledge the robustness of this approach across various scenarios, including cross-domain challenges, model shifts, and adversarial attacks.”
From A Survey on LLM-Generated Text Detection: Necessity, Methods, and Future Directions (2025) · cites this paper as Tulchinskii et al. 2023
Tulchinskii et al. (2023) approach's reliability falters when confronted with suboptimal or high-temperature generators.
“However, its reliability falters when confronted with suboptimal or high-temperature generators.”
From A Survey on LLM-Generated Text Detection: Necessity, Methods, and Future Directions (2025) · cites this paper as Tulchinskii et al. 2023
The Overview describes the intrinsic-dimension detector as robust, and the Method reports detection results for several generators. Neither section addresses whether its reliability declines specifically with suboptimal or high-temperature generators (searched: high-temperature generators, sampling temperature, generator quality). (LLM, conf=0.72)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
GPT detectors are biased against non-native English writers
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023) · Patterns · doi:10.1016/j.patter.2023.100779 · Cited by 673 · cited 2× · see citing works
“In order to assess the bias in our estimator, we use the data provided by Liang et al. [2023].”
BackgroundMethodA Watermark for Large Language Models
Kirchenbauer, J., Geiping, J., Wen, Y., Katz, J., Miers, I., & Goldstein, T. (2023) · arXiv · doi:10.48550/arxiv.2301.10226 · Cited by 194 · cited 1× · see citing works
“Another idea, known as watermarking , is to inject some detectable artifacts into model generations; for instance, Kirchenbauer et al. [2023] propose to intentionally inject a statistical skew that can be detected in a text sample.”
BackgroundCan AI-Generated Text be Reliably Detected?
Sadasivan, V. S., Kumar, A., Balasubramanian, S., Wang, W., & Feizi, S. (2023) · arXiv · doi:10.48550/arxiv.2303.11156 · Cited by 161 · see citing works
Paraphrasing evades detectors of AI-generated text, but retrieval is an effective defense
Krishna, K., Song, Y., Karpinska, M., Wieting, J., & Iyyer, M. (2023) · arXiv · doi:10.48550/arxiv.2303.13408 · Cited by 150 · cited 8× · see citing works
“Data for Wikipedia and Reddit was taken from [Krishna et al., 2023]; as for StackExchange, we assembled the dataset ourselves following the same scheme as Krishna et al. [2023].”
BackgroundMethodInfluential
Flagged first, then by citations
Research works printed but not linked: 1
- F. Chazal and B. Michel. An introduction to topological data analysis: fundamental and practical aspects for data scientists. Frontiers in Artificial Intelligence, 4:667963, Sep 2021. doi: 10.3389/frai.2021.667963 · DOI
Cited by
Sorted by citations
A Survey on LLM-Generated Text Detection: Necessity, Methods, and Future Directions
Wu, J., Yang, S., Zhan, R., Yuan, Y., Chao, L. S., & Wong, D. F. (2025) · Computational Linguistics · doi:10.1162/coli_a_00549 · Cited by 212 · see citing works
Quantifying large language model usage in scientific papers
Liang, W., Zhang, Y., Wu, Z., Lepp, H., Ji, W., Zhao, X., Cao, H., Liu, S., He, S., Huang, Z., et al. (2025) · Nature Human Behaviour · doi:10.1038/s41562-025-02273-8 · Cited by 80 · see citing works
Liang, W., Izzo, Z., Zhang, Y., Lepp, H., Cao, H., Zhao, X., Chen, L., Ye, H., Liu, S., Huang, Z., et al. (2024) · arXiv · doi:10.48550/arxiv.2403.07183 · Cited by 73 · see citing works
“Tulchinskii et al. (2023) show that machine text has lower intrinsic dimensionality than human writing, as measured by persistent homology for dimension estimation.”
Spotting LLMs With Binoculars: Zero-Shot Detection of Machine-Generated Text
Hans, A., Schwarzschild, A., Cherepanova, V., Kazemi, H., Saha, A., Goldblum, M., Geiping, J., & Goldstein, T. (2024) · arXiv · doi:10.48550/arxiv.2401.12070 · Cited by 24 · see citing works
Enhancing the Robustness of AI-Generated Text Detectors: A Survey
Liu, X., Li, Y., & Li, K. (2025) · Mathematics · doi:10.3390/math13132145 · Cited by 10 · see citing works
- Authors
- Tulchinskii, E., Kuznetsov, K., Kushnareva, L., Cherniavskii, D., Barannikov, S., Piontkovskaya, I., Nikolenko, S., & Burnaev, E.
- Venue
- arXiv
- Publisher
- arXiv
- Published
- 2023
- Pages
- 39257-39276
- ISSN
- 1049-5258 (ISSN-L)
- arXiv
- 2306.04723
- OpenAlex
- W7133226946
- Cited by
- 43
- References
- 4
- Citations by year
- 2026: 7 · 2025: 19 · 2024: 12 · 2023: 5
- FWCI
- 7.63
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- neural information processing systems (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- HAL (submittedVersion, other-oa)
- arXiv (Cornell University) (cc-by)
- Funders
- Russian Foundation for Basic Research (21-51-12005 NNIO_a)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Authorship Attribution and Profiling
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Keywords
- AI-generated text detection, intrinsic dimension estimation, deepfakes, generative AI
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
