REVIEW 15 cited by
DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
The increasing fluency and widespread usage of large language models (LLMs) highlight the desirability of corresponding tools aiding detection of LLM-generated text. In this paper, we identify a property of the structure of an LLM's probability function that is useful for such detection. Specifically, we demonstrate that text sampled from an LLM tends to occupy negative curvature regions of the model's log probability function. Leveraging this observation, we then define a new curvature-based criterion for judging if a passage is generated from a given LLM. This approach, which we call DetectGPT, does not require training a separate classifier, collecting a dataset of real or generated passages, or explicitly watermarking generated text. It uses only log probabilities computed by the model of interest and random perturbations of the passage from another generic pre-trained language model (e.g., T5). We find DetectGPT is more discriminative than existing zero-shot methods for model sample detection, notably improving detection of fake news articles generated by 20B parameter GPT-NeoX from 0.81 AUROC for the strongest zero-shot baseline to 0.95 AUROC for DetectGPT. See https://ericmitchell.ai/detectgpt for code, data, and other project information.
Forward citations
Cited by 15 Pith papers
-
Tailored untruths: How personalisation challenges LLM safeguards
A 1.6-million-text study of eight LLMs in four languages finds that adding demographic personae to disinformation prompts raises jailbreak rates from 78% to 82%.
-
Attacks on Machine-Text Detectors Retain Stylistic Fingerprints
A style-aware paraphrasing attack evades all nine tested AI-text detectors at the single-document level, but multi-document analysis makes the attack detectable again.
-
Policies Permitting LLM Use for Polishing Peer Reviews Are Currently Not Enforceable
State-of-the-art AI detectors misclassify a non-trivial fraction of LLM-polished peer reviews as fully AI-generated, rendering polishing-only policies currently unenforceable.
-
Using Large Language Models for Idea Generation in Innovation
GPT-4-generated product ideas had higher average purchase intent than student ideas and made up 35 of the top 40 ideas, while being rated less novel and more similar to each other.
-
AI's Capability in Assisting Scientific Research in Physics, Astrophysics, and Cosmology I: Literature Review
In a controlled test, three mid-2025 LLMs shared under 6% of literature references with physics experts, and 64% of their real references had at least one metadata error.
-
Artificial Epanorthosis: Why large language models overuse a classical rhetorical figure, and how to mitigate it
LLMs overuse the 'not X, but Y' self-correction pattern in persuasive registers and underuse it in informal Q&A; a prompt or a detachable LoRA dial adjusts it to human levels.
-
Black-Box Detection of LLM-Generated Text Using Generalized Jensen-Shannon Divergence
SurpMark detects machine-generated text by estimating state-transition matrices from discretized surprisals and scoring them with generalized Jensen-Shannon divergence to human versus machine references.
-
Optimizing Token Choice for Code Watermarking: An RL Approach
An RL-trained policy adaptively biases token choices to watermark LLM-generated code while preserving executable behavior.
-
Detecting LLM-generated Code with Subtle Modification by Adversarial Training
CodeGPTSensor+, trained with adversarial samples that combine identifier renaming and structure transformation, is substantially more robust to subtle modifications of LLM-generated code than the original CodeGPTSensor.
-
Detecting AI Assistance in Abstract Complex Tasks
Converting behavioral search traces into image channels plus an exploration/exploitation time series lets a small CNN-RNN detect AI assistance with about 86% accuracy on a balanced lab task.
-
I Know Which LLM Wrote Your Code Last Summer: LLM generated Code Stylometry for Authorship Attribution
A fine-tuned encoder-only CodeT5 model attributes LLM-generated C code to its source model with up to 97.56% binary and 95.40% five-class accuracy on a new 32,000-program benchmark.
-
AI's Capability in Assisting Scientific Research in Physics, Astrophysics, and Cosmology II: Project Planning and Proposal Evaluation
AI-generated one-page research proposals are scored about the same as human-written ones by human reviewers, but AI reviewers favor AI-written proposals by roughly one point and detect authorship perfectly.
-
A Comprehensive Dataset for Human vs. AI Generated Text Detection
A dataset of ~58k NYT articles plus AI rewrites from six LLMs, evaluated with a rewrite-distance baseline reaching 58.35% detection and 8.92% attribution accuracy.
-
Can You Detect the Difference?
A 2,000-sample comparison finds diffusion-generated LLaDA text can match human perplexity and burstiness when rephrasing, while LLaMA text is more predictable and easier to flag.
-
AI-Generated Content in Cross-Domain Applications: Research Trends, Challenges and Propositions
A cross-domain vision paper that surveys AI-generated content and proposes research directions, without introducing new empirical results.
Discussion (0). Sign in to comment.