{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"awesome-detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","slug":"detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","name":"DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature (DetectGPT)","type":"product","url":"https://arxiv.org/abs/2301.11305","page_url":"https://unfragile.ai/detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","categories":["productivity"],"tags":[],"pricing":{"model":"unknown","free":false,"starting_price":null},"status":"inactive","verified":false},"capabilities":[{"id":"awesome-detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt__cap_0","uri":"capability://safety.moderation.zero.shot.machine.generated.text.detection.via.probability.curvature.analysis","name":"zero-shot machine-generated text detection via probability curvature analysis","description":"Detects machine-generated text without requiring training data by analyzing the curvature of token probability distributions from a reference language model. The method computes the difference between log-probabilities assigned by the reference model to original text versus perturbed text (with randomly masked tokens replaced), measuring how sharply probability distributions change. This probability curvature signature distinguishes human-written text (which exhibits different distributional properties) from LLM-generated text without fine-tuning or labeled datasets.","intents":["Detect whether a given text passage was generated by an LLM or written by a human without access to training data","Identify machine-generated content in academic submissions, user-generated content platforms, or content moderation pipelines","Evaluate the authenticity of text without requiring labeled examples of machine-generated vs human text"],"best_for":["Content moderation teams needing zero-shot detection without labeled training data","Academic integrity systems detecting AI-generated essays or papers","Researchers studying LLM detection methods and probability-based approaches"],"limitations":["Requires access to a reference language model (e.g., GPT-2, GPT-3) for probability computation, adding computational overhead","Detection accuracy degrades when text is heavily paraphrased or edited after generation","Assumes the reference model has reasonable coverage of the language domain; may fail on specialized or non-English text","Probability curvature signal may weaken as LLMs improve and generate more human-like distributions","No built-in handling for multi-language or domain-specific text without model retraining"],"requires":["Access to a pre-trained language model (GPT-2, GPT-3, or similar) with probability output capabilities","Text input of sufficient length (typically 100+ tokens for reliable detection)","Computational resources to run forward passes through the reference model for each detection query"],"input_types":["text (plain text passages, documents, essays)"],"output_types":["binary classification (machine-generated vs human-written)","confidence score (probability curvature magnitude)","probability distribution analysis"],"categories":["safety-moderation","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt__cap_1","uri":"capability://data.processing.analysis.masked.token.perturbation.for.probability.distribution.sampling","name":"masked token perturbation for probability distribution sampling","description":"Generates perturbed versions of input text by randomly masking tokens and replacing them with samples from the reference model's probability distribution. For each masked position, the method samples alternative tokens according to the model's predicted probabilities, creating multiple variants of the original text. This perturbation strategy allows the detector to measure how probability distributions shift when text is modified, providing the signal for curvature-based detection without requiring explicit training on synthetic data.","intents":["Generate multiple plausible variations of input text to measure probability distribution stability","Create contrastive examples that expose differences in how reference models assign probabilities to human vs machine text","Compute statistical signatures of text authenticity through systematic perturbation"],"best_for":["Researchers analyzing probability distributions of language models","Detection systems that need to generate contrastive examples on-the-fly without pre-computed datasets"],"limitations":["Sampling from probability distributions introduces stochasticity; results vary across runs unless seeded","Computational cost scales with number of perturbations and text length (each perturbation requires a forward pass)","Quality of perturbations depends on the reference model's probability calibration; poorly calibrated models produce unreliable variants","May not preserve semantic meaning of original text, potentially introducing artifacts"],"requires":["Reference language model with token-level probability outputs","Masking mechanism compatible with the model architecture (e.g., [MASK] token for BERT-style models)","Sampling strategy (temperature-based or top-k/top-p) for controlling perturbation diversity"],"input_types":["text (tokenized or raw)"],"output_types":["perturbed text variants","probability scores for sampled tokens","distribution of alternative token choices"],"categories":["data-processing-analysis","text-generation-language"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt__cap_2","uri":"capability://safety.moderation.reference.model.agnostic.detection.scoring.with.cross.model.compatibility","name":"reference model-agnostic detection scoring with cross-model compatibility","description":"Computes detection scores using any pre-trained language model as a reference, without requiring the reference model to be the same model that generated the suspect text. The method calculates probability curvature relative to the reference model's distribution, enabling detection even when the generating model is unknown or proprietary. This architecture allows deployment with readily available models (e.g., GPT-2, open-source LLMs) while detecting text from any LLM, including closed-source systems.","intents":["Detect text generated by proprietary LLMs (e.g., GPT-3, GPT-4) using only open-source reference models","Deploy detection systems without knowing which LLM generated the suspect text","Switch reference models at inference time to adapt to different domains or improve detection accuracy"],"best_for":["Content moderation platforms that need to detect text from multiple LLM sources","Organizations without access to the generating LLM but with access to open-source alternatives","Research teams studying cross-model detection generalization"],"limitations":["Detection accuracy may vary depending on the reference model chosen; some models are better detectors than others","Requires the reference model to have reasonable probability calibration; miscalibrated models produce unreliable scores","No guarantee of detection if the reference model and generating model have very different architectures or training objectives","Computational cost depends on reference model size; larger models provide better detection but slower inference"],"requires":["At least one pre-trained language model with probability output capabilities","Ability to run the reference model on the same hardware or via API","Text input of sufficient length for reliable probability estimation"],"input_types":["text (plain text passages)"],"output_types":["detection score (probability curvature magnitude)","confidence threshold for binary classification","per-token probability contributions to overall score"],"categories":["safety-moderation","data-processing-analysis"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"awesome-detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt__cap_3","uri":"capability://data.processing.analysis.probability.curvature.computation.with.statistical.significance.testing","name":"probability curvature computation with statistical significance testing","description":"Calculates a numerical score representing the curvature of token probability distributions by measuring the divergence between log-probabilities of original and perturbed text. The method computes statistics such as the mean and variance of probability differences across tokens, enabling statistical significance testing to distinguish genuine machine-generated text from natural variation in human writing. This statistical framework provides both a point estimate (curvature score) and confidence intervals for detection decisions.","intents":["Quantify the statistical signature of machine-generated text in a way that enables threshold-based classification","Assess the confidence or uncertainty of detection decisions through statistical testing","Compare detection scores across different texts or models using a standardized metric"],"best_for":["Systems requiring explainable detection scores with statistical justification","Researchers analyzing the statistical properties of LLM-generated text","Applications needing confidence intervals or p-values for detection decisions"],"limitations":["Statistical significance depends on text length; short texts may not provide sufficient statistical power","Assumes independence of token probabilities, which may not hold for correlated language patterns","Threshold selection for binary classification requires calibration on validation data, reducing true zero-shot properties","Statistical tests assume specific distributions (e.g., normality) that may not hold for probability curvature scores"],"requires":["Multiple perturbed text samples to compute variance and statistical measures","Sufficient text length (typically 100+ tokens) for reliable statistical estimation","Optional: validation data for threshold calibration"],"input_types":["probability distributions (log-probabilities from reference model)","text variants (original and perturbed)"],"output_types":["curvature score (numerical)","statistical significance (p-value or confidence interval)","per-token probability differences"],"categories":["data-processing-analysis","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":21,"verified":false,"data_access_risk":"low","permissions":["Access to a pre-trained language model (GPT-2, GPT-3, or similar) with probability output capabilities","Text input of sufficient length (typically 100+ tokens for reliable detection)","Computational resources to run forward passes through the reference model for each detection query","Reference language model with token-level probability outputs","Masking mechanism compatible with the model architecture (e.g., [MASK] token for BERT-style models)","Sampling strategy (temperature-based or top-k/top-p) for controlling perturbation diversity","At least one pre-trained language model with probability output capabilities","Ability to run the reference model on the same hardware or via API","Text input of sufficient length for reliable probability estimation","Multiple perturbed text samples to compute variance and statistical measures"],"failure_modes":["Requires access to a reference language model (e.g., GPT-2, GPT-3) for probability computation, adding computational overhead","Detection accuracy degrades when text is heavily paraphrased or edited after generation","Assumes the reference model has reasonable coverage of the language domain; may fail on specialized or non-English text","Probability curvature signal may weaken as LLMs improve and generate more human-like distributions","No built-in handling for multi-language or domain-specific text without model retraining","Sampling from probability distributions introduces stochasticity; results vary across runs unless seeded","Computational cost scales with number of perturbations and text length (each perturbation requires a forward pass)","Quality of perturbations depends on the reference model's probability calibration; poorly calibrated models produce unreliable variants","May not preserve semantic meaning of original text, potentially introducing artifacts","Detection accuracy may vary depending on the reference model chosen; some models are better detectors than others","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.05,"quality":0.23,"ecosystem":0.25,"match_graph":0.25,"freshness":0.5,"weights":{"adoption":0.25,"quality":0.25,"ecosystem":0.1,"match_graph":0.35,"freshness":0.05}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"inactive","updated_at":"2026-06-17T09:51:03.038Z","last_scraped_at":"2026-05-03T14:00:27.894Z","last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","compare_url":"https://unfragile.ai/compare?artifact=detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt"}},"signature":"seiIC8UvhcSxvHadmsO8+qdzK+nAJBFVyblbXNNaQ7qBLZPCe3gQ4f1QomvreDECerwYLXhso9lMSJ2UHhDtCw==","signedAt":"2026-06-18T17:16:02.968Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","artifact":"https://unfragile.ai/detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","verify":"https://unfragile.ai/api/v1/verify?slug=detectgpt-zero-shot-machine-generated-text-detection-using-probability-curvature-detectgpt","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}