---
title: "Zero-Shot Detection of LLM-Generated Text using Temperature"
description: "The proposed detector, NTS, thresholds a normalized TS computed on LogPPL. 27 references."
authors: ["Shixuan Ma", "Jiahao Li", "Zhendong Mao", "Quan Wang"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.1748/"
doi: "10.18653/v1/2026.acl-long.1748"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/ma-2026-zero-shot-detection-llm-generated/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ma-2026-zero-shot-detection-llm-generated/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/48/) / Zero-Shot Detection of LLM-Generated Text using Temperature Sensitivity

# Zero-Shot Detection of LLM-Generated Text using Temperature Sensitivity

**Authors:** Shixuan Ma, [Jiahao Li](https://md.research.elamur.ai/free/en/authors/jiahao-li/), [Zhendong Mao](https://md.research.elamur.ai/free/en/authors/zhendong-mao/) and [Quan Wang](https://md.research.elamur.ai/free/en/authors/quan-wang/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.1748/) · [PDF](https://aclanthology.org/2026.acl-long.1748.pdf) · [DOI 10.18653/v1/2026.acl-long.1748](https://doi.org/10.18653/v1/2026.acl-long.1748) · [OpenAlex](https://openalex.org/W7166888511)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Zero-shot detectors of machine-written text usually score a passage with a single surrogate-model statistic, such as LogPPL or Entropy. The authors argue that these aggregate numbers compress the token distributions too much. As a result, human-written and LLM-generated texts can get almost the same values even when their underlying distributions differ.

Their alternative is to vary the surrogate model's decoding temperature and measure how much a statistic shifts. They call this shift Temperature Sensitivity (TS). The reasoning is that LLM output tends to fall in the high-probability head of the distribution, where log-probabilities respond more strongly to temperature than in the middle region where human word choices usually sit.

The proposed detector, NTS, thresholds a normalized TS computed on LogPPL. It needs one forward pass through one surrogate model. In tests across three datasets, five languages, over 15 generators and more than 12 domains, NTS beats strong zero-shot and supervised baselines and stays stable across temperature settings. The work matters because it gives a cheap, training-free signal for identifying LLM text that generalizes well.

## Key points
- TS is defined as the absolute gap in a statistic (e.g., LogPPL) between low and high decoding temperature, and LLM-generated text shows consistently higher TS. Code is available at github.com/Shixuan-Ma/NTS.
- NTS was compared with 12 baselines (7 zero-shot, e.g., IRM, Binoculars and Fast-DetectGPT; 5 supervised) on datasets including HART and RAID.
- A Random Forest trained on TS features reached AUROC 0.928 on RAID and 0.924 on HART, ahead of Biscope (0.903 and 0.907), Ghostbuster, IMBD, RADAR and RoBERTa.
- On RAID, NTS had the highest Mean Confidence (average 0.340), compared with Fast-DetectGPT (0.308) and Binoculars (0.301).
- Large base surrogates such as Falcon-7B and Llama models give AUROC above 0.9, while small or instruction-tuned surrogates perform worse. Performance also drops under high-temperature perturbations.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Science and Technology of China](https://md.research.elamur.ai/free/en/universities/university-of-science-and-technology-of-china/) — University, Hefei, China · 108 works · 112,698 citations
- [Beijing University of Posts and Telecommunications](https://md.research.elamur.ai/free/en/universities/beijing-university-of-posts-and-telecommunications/) — University, Beijing, China · 57 works · 32,843 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works

## Cite

Cite as: Shixuan Ma, Jiahao Li, Zhendong Mao and Quan Wang. 2026. Zero-Shot Detection of LLM-Generated Text using Temperature Sensitivity. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{ma2026zero,
  title = {{Zero-Shot Detection of LLM-Generated Text using Temperature Sensitivity}},
  author = {Shixuan Ma, Jiahao Li, Zhendong Mao and Quan Wang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1748},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1748}
}
```

- [← previous Zero-Shot Context-Aware ASR for Diverse Arabic Varieties](https://md.research.elamur.ai/free/en/papers/talafha-2026-zero-shot-context-aware-asr/)
- [next → ZIP: Quantifying Which Words Matter in Zero-Shot Instructional Prompts](https://md.research.elamur.ai/free/en/papers/sadr-2026-zip-quantifying-which-words-matter/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/48/)

## References

27 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Alshamsi, Alessandro Cappelli, Ruxandra Cojocaru, Mérouane Debbah, Étienne Goffinet, Daniel Hesslow, Julien Launay, Quentin Malartic, Daniele Mazzotta, Badreddine Noune, Baptiste Pannier, and Guilherme Penedo. 2023. [The falcon series of open language mo…](https://md.research.elamur.ai/free/en/papers/almazrouei-2023-falcon-series-open-language-models/)
2. Guangsheng Bao, Lihua Rong, Yanbin Zhao, Qiji Zhou, and Yue Zhang. 2025. Decoupling content and expression: Two-dimensional detection of ai-generated text. Preprint, arXiv:2503.00258. — [arXiv:2503.00258](https://arxiv.org/abs/2503.00258) [find](https://scholar.google.com/scholar?q=Guangsheng+Bao%2C+Lihua+Rong%2C+Yanbin+Zhao%2C+Qiji+Zhou%2C+and+Yue+Zhang%2E+2025%2E+Decoupling+content+and+expression%3A+Two%2Ddimensional+detection+of+ai%2Dgenerated+text%2E+Preprint%2C+arXiv%3A2503%2E00258%2E+%E2%80%94+%5BarXiv%3A2503%2E00)
3. Guangsheng Bao, Yanbin Zhao, Zhiyang Teng, Linyi Yang, and Yue Zhang. 2024. [Fast-detectgpt: Efficient zero-shot detection of machine-generated text via conditional probability curvature](https://md.research.elamur.ai/free/en/papers/bao-2023-fast-detectgpt-efficient-zero-shot/). Preprint, arXiv:2310.05130.
4. Erik Brynjolfsson, Danielle Li, and Lindsey Raymond. 2025. Generative ai at work*. The Quarterly Journal of Economics, 140(2):889–942. [find](https://scholar.google.com/scholar?q=Erik+Brynjolfsson%2C+Danielle+Li%2C+and+Lindsey+Raymond%2E+2025%2E+Generative+ai+at+work%2A%2E+The+Quarterly+Journal+of+Economics%2C+140%282%29%3A889%E2%80%93942%2E)
5. Jiaqi Chen, Xiaoye Zhu, Tianyang Liu, Ying Chen, Xinhui Chen, Yiwen Yuan, Chak Tou Leong, Zuchao Li, Tang Long, Lei Zhang, Chenyu Yan, Guanghao Mei, Jie Zhang, and Lefei Zhang. 2024. Imitate before detect: Aligning machine stylistic preference for machine-revised text detection. Preprint, arXiv:2412… — [arXiv:2412.10432](https://arxiv.org/abs/2412.10432) [find](https://scholar.google.com/scholar?q=Jiaqi+Chen%2C+Xiaoye+Zhu%2C+Tianyang+Liu%2C+Ying+Chen%2C+Xinhui+Chen%2C+Yiwen+Yuan%2C+Chak+Tou+Leong%2C+Zuchao+Li%2C+Tang+Long%2C+Lei+Zhang%2C+Chenyu+Yan%2C+Guanghao+Mei%2C+Jie+Zhang%2C+and+Lefei+Zhang%2E+2024%2E+Imitate+before+de)
6. Liam Dugan, Alyssa Hwang, Filip Trhlík, Andrew Zhu, Josh Magnus Ludan, Hainiu Xu, Daphne Ippolito, and Chris Callison-Burch. 2024. [RAID: A shared benchmark for robust evaluation of machine-generated text detectors](https://md.research.elamur.ai/free/en/papers/dugan-2024-raid-shared-benchmark-robust-evaluation/). In Proceedings of the 62nd Annual Meeting of the Association for Computational Lingui…
7. Damian Okaibedi Eke. 2023. Chatgpt and the rise of generative ai: Threat to academic integrity? Journal of Responsible Technology, 13:100060. [find](https://scholar.google.com/scholar?q=Damian+Okaibedi+Eke%2E+2023%2E+Chatgpt+and+the+rise+of+generative+ai%3A+Threat+to+academic+integrity%3F+Journal+of+Responsible+Technology%2C+13%3A100060%2E)
8. Gemini Team Google. 2023. [Gemini: A family of highly capable multimodal models](https://md.research.elamur.ai/free/en/papers/gemini-team-2023-gemini-family-highly-capable-multimodal/). arXiv preprint arXiv:2312.11805.
9. Biyang Guo, Xin Zhang, Ziyuan Wang, Minqi Jiang, Jinran Nie, Yuxuan Ding, Jianwei Yue, and Yupeng Wu. 2023. [How close is chatgpt to human experts? comparison corpus, evaluation, and detection](https://md.research.elamur.ai/free/en/papers/guo-2023-close-chatgpt-human-experts-comparison/). Preprint, arXiv:2301.07597.
10. Hanxi Guo, Siyuan Cheng, Xiaolong Jin, Zhuo Zhang, Kaiyuan Zhang, Guanhong Tao, Guangyu Shen, and Xiangyu Zhang. 2024. Biscope: Ai-generated text detection by checking memorization of preceding tokens. Advances in Neural Information Processing Systems (NeurIPS), 37:104065–104090. [find](https://scholar.google.com/scholar?q=Hanxi+Guo%2C+Siyuan+Cheng%2C+Xiaolong+Jin%2C+Zhuo+Zhang%2C+Kaiyuan+Zhang%2C+Guanhong+Tao%2C+Guangyu+Shen%2C+and+Xiangyu+Zhang%2E+2024%2E+Biscope%3A+Ai%2Dgenerated+text+detection+by+checking+memorization+of+preceding+tokens)
11. Abhimanyu Hans, Avi Schwarzschild, Valeriia Cherepanova, Hamid Kazemi, Aniruddha Saha, Micah Goldblum, Jonas Geiping, and Tom Goldstein. 2024. Spotting llms with binoculars: Zero-shot detection of machine-generated text. Preprint, arXiv:2401.12070. [find](https://scholar.google.com/scholar?q=Abhimanyu+Hans%2C+Avi+Schwarzschild%2C+Valeriia+Cherepanova%2C+Hamid+Kazemi%2C+Aniruddha+Saha%2C+Micah+Goldblum%2C+Jonas+Geiping%2C+and+Tom+Goldstein%2E+2024%2E+Spotting+llms+with+binoculars%3A+Zero%2Dshot+detection+of+mac)
12. Tatsunori B. Hashimoto, Hugh Zhang, and Percy Liang. 2019. Unifying human and statistical evaluation for natural language generation. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and… [find](https://scholar.google.com/scholar?q=Tatsunori+B%2E+Hashimoto%2C+Hugh+Zhang%2C+and+Percy+Liang%2E+2019%2E+Unifying+human+and+statistical+evaluation+for+natural+language+generation%2E+In+Proceedings+of+the+2019+Conference+of+the+North+American+Chapte)
13. Hello-SimpleAI. 2023. chatgpt-detector-roberta (revision d2b342c). [find](https://scholar.google.com/scholar?q=Hello%2DSimpleAI%2E+2023%2E+chatgpt%2Ddetector%2Droberta+%28revision+d2b342c%29%2E)
14. Xiaomeng Hu, Pin-Yu Chen, and Tsung-Yi Ho. 2023. Radar: Robust ai-text detection via adversarial learning. In Advances in Neural Information Processing Systems, volume 36, pages 15077–15095. Curran Associates, Inc. [find](https://scholar.google.com/scholar?q=Xiaomeng+Hu%2C+Pin%2DYu+Chen%2C+and+Tsung%2DYi+Ho%2E+2023%2E+Radar%3A+Robust+ai%2Dtext+detection+via+adversarial+learning%2E+In+Advances+in+Neural+Information+Processing+Systems%2C+volume+36%2C+pages+15077%E2%80%9315095%2E+Curran+As)
15. Junjie Huang, Ruiquan Zhang, Jinsong Su, and Yidong Chen. 2025. Tempparaphraser:“heating up” text to evade ai-text detection through paraphrasing. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 31542– 31561. [find](https://scholar.google.com/scholar?q=Junjie+Huang%2C+Ruiquan+Zhang%2C+Jinsong+Su%2C+and+Yidong+Chen%2E+2025%2E+Tempparaphraser%3A%E2%80%9Cheating+up%E2%80%9D+text+to+evade+ai%2Dtext+detection+through+paraphrasing%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Me)
16. Runheng Liu, Heyan Huang, Xingchen Xiao, and Zhijing Wu. 2025. Zero-shot detection of llm-generated text via implicit reward model. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Runheng+Liu%2C+Heyan+Huang%2C+Xingchen+Xiao%2C+and+Zhijing+Wu%2E+2025%2E+Zero%2Dshot+detection+of+llm%2Dgenerated+text+via+implicit+reward+model%2E+In+The+Thirty%2Dninth+Annual+Conference+on+Neural+Information+Processi)
17. Eric Mitchell, Yoonho Lee, Alexander Khazatsky, Christopher D. Manning, and Chelsea Finn. 2023. [Detectgpt: Zero-shot machine-generated text detection using probability curvature](https://md.research.elamur.ai/free/en/papers/mitchell-2023-detectgpt-zero-shot-machine-generated/). Preprint, arXiv:2301.11305.
18. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, and 262 others.… — [GPT-4 Technical Report](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/)
19. Yikang Pan, Liangming Pan, Wenhu Chen, Preslav Nakov, Min-Yen Kan, and William Yang Wang. 2023. [On the risk of misinformation pollution with large language models](https://md.research.elamur.ai/free/en/papers/pan-2023-risk-misinformation-pollution-large-language/). Preprint, arXiv:2305.13661.
20. Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. 2023. The impact of ai on developer productivity: Evidence from github copilot. Preprint, arXiv:2302.06590. — [arXiv:2302.06590](https://arxiv.org/abs/2302.06590) [find](https://scholar.google.com/scholar?q=Sida+Peng%2C+Eirini+Kalliamvakou%2C+Peter+Cihon%2C+and+Mert+Demirer%2E+2023%2E+The+impact+of+ai+on+developer+productivity%3A+Evidence+from+github+copilot%2E+Preprint%2C+arXiv%3A2302%2E06590%2E+%E2%80%94+%5BarXiv%3A2302%2E06590%5D%28https%3A%2F%2F)
21. Jiameng Pu, Zain Sarwar, Sifat Muhammad Abdullah, Abdullah Rehman, Yoonjin Kim, Parantapa Bhattacharya, Mobin Javed, and Bimal Viswanath. 2022. Deepfake text detection: Limitations and opportunities. Preprint, arXiv:2210.09421. — [arXiv:2210.09421](https://arxiv.org/abs/2210.09421) [find](https://scholar.google.com/scholar?q=Jiameng+Pu%2C+Zain+Sarwar%2C+Sifat+Muhammad+Abdullah%2C+Abdullah+Rehman%2C+Yoonjin+Kim%2C+Parantapa+Bhattacharya%2C+Mobin+Javed%2C+and+Bimal+Viswanath%2E+2022%2E+Deepfake+text+detection%3A+Limitations+and+opportunities%2E+)
22. Shushanta Pudasaini, Luis Miralles-Pechuán, David Lil-lis, and Marisa Llorens Salvador. 2024. Survey on plagiarism detection in large language models: The impact of chatgpt and gemini on academic integrity. Preprint, arXiv:2407.13105. — [arXiv:2407.13105](https://arxiv.org/abs/2407.13105) [find](https://scholar.google.com/scholar?q=Shushanta+Pudasaini%2C+Luis+Miralles%2DPechu%C3%A1n%2C+David+Lil%2Dlis%2C+and+Marisa+Llorens+Salvador%2E+2024%2E+Survey+on+plagiarism+detection+in+large+language+models%3A+The+impact+of+chatgpt+and+gemini+on+academic+inte)
23. Sara Sabour, Nicholas Frosst, and Geoffrey E Hinton. 2017. Dynamic routing between capsules. Preprint, arXiv:1710.09829. — [arXiv:1710.09829](https://arxiv.org/abs/1710.09829) [find](https://scholar.google.com/scholar?q=Sara+Sabour%2C+Nicholas+Frosst%2C+and+Geoffrey+E+Hinton%2E+2017%2E+Dynamic+routing+between+capsules%2E+Preprint%2C+arXiv%3A1710%2E09829%2E+%E2%80%94+%5BarXiv%3A1710%2E09829%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1710%2E09829%29)
24. Irene Solaiman, Miles Brundage, Jack Clark, Amanda Askell, Ariel Herbert-Voss, Jeff Wu, Alec Radford, Gretchen Krueger, Jong Wook Kim, Sarah Kreps, and 1 others. 2019. [Release strategies and the social impacts of language models](https://md.research.elamur.ai/free/en/papers/solaiman-2019-release-strategies-social-impacts-language/). arXiv preprint arXiv:1908.09203.
25. Jinyan Su, Terry Zhuo, Di Wang, and Preslav Nakov. 2023. [DetectLLM: Leveraging log rank information for zero-shot detection of machine-generated text](https://md.research.elamur.ai/free/en/papers/su-2023-detectllm-leveraging-log-rank-information/). In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 12395–12412, Singapore. Association for Computational Linguistics.
26. Ruixiang Tang, Yu-Neng Chuang, and Xia Hu. 2024. The science of detecting llm-generated text. Commun. ACM, 67(4):50–59. [find](https://scholar.google.com/scholar?q=Ruixiang+Tang%2C+Yu%2DNeng+Chuang%2C+and+Xia+Hu%2E+2024%2E+The+science+of+detecting+llm%2Dgenerated+text%2E+Commun%2E+ACM%2C+67%284%29%3A50%E2%80%9359%2E)
27. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. [Llama: Open and efficient foundation language models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-open-efficient-foundation-language/). Preprint…

*Updated 1 November 2026*