People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text
Paper • 2501.15654 • Published • 16
Datasets:
RADAR float64 0 0.95 | annotator_1 dict | annotator_2 dict | annotator_3 dict | annotator_4 dict | annotator_5 dict | article stringlengths 2.48k 7.24k | author stringlengths 8 30 | binoculars_accuracy stringclasses 2
values | binoculars_lowfpr stringclasses 2
values | e5_lora float64 0.04 0.9 | expert_majority_vote stringclasses 2
values | generation_model stringclasses 5
values | gptzero stringlengths 8.19k 25k | ground_truth stringclasses 2
values | id int64 3 56 | issue stringlengths 6 22 | issue_original_type stringclasses 2
values | link stringlengths 55 175 | majority_vote stringclasses 2
values | pangram stringlengths 2.71k 7.61k | pangram_humanizers stringlengths 2.71k 7.51k | prompt_id int64 4 26 | section stringlengths 4 26 | source stringclasses 7
values | source_dataset stringclasses 1
value | source_row_index int64 3 295 | source_sha256 stringclasses 1
value | source_url stringclasses 1
value | split stringclasses 1
value | split_method stringclasses 1
value | split_seed stringclasses 1
value | sub-title stringlengths 42 216 | title stringlengths 28 124 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
0.352621 | {
"comment": "The quotes in this piece of text come across as very natural. Some fit in well, others need surrounding text to make sense, and the sources are varied. The professors are also not referred to as doctors, even though they're probably qualified enough. There's also just an absence of anything that might p... | {
"comment": "There are none of the usual AI-repeated words. There is a wider range of vocabulary than usual for AI.",
"confidence": 4,
"guess": "Human-Generated"
} | {
"comment": "Here's why I think it's human-generated: Instead of saying things like 'researchers don't agree on' AI would have said 'researchers disagree on'. Missing punctuation marks. ",
"confidence": 5,
"guess": "Human-Generated"
} | {
"comment": "While the writing is more simplistic, it's able to convey the topic well. It uses dashes, colons, and commas to intersperse information with quotes, works with simple, easy to understand phrases such as \"a leg up compared to all other species.\" and \"Pagel, on the other hand, is less sure about hand g... | {
"comment": "Extensive use of personal pronouns. Quotations provide detail. Distinct variation in sentence and paragraph length. Highlighted sentence, for e.g., is 40 words. Rare in machine-generated text territory. Use of idiom in introductory sentence.",
"confidence": 5,
"guess": "Human-Generated"
} | Language gave humans a leg up compared to all other species. By working together outside of the family unit through spoken — and later written — communication, we were able to hunt and gather, farm, and build great civilizations.
There’s a lot we don’t know about how it all started, though. That’s because there were n... | Sara Novak | Most likely human-generated | Most likely human-generated | 0.149416 | Human-Generated | gpt-4o | {'version': '2025-01-09-base', 'scanId': '63323f35-3191-4361-8bf3-7c72323e8c58', 'documents': [{'sentences': [{'generated_prob': 0.0003501292085275054, 'sentence': 'Language gave humans a leg up compared to all other species.', 'perplexity': 0, 'highlight_sentence_for_ai': False}, {'generated_prob': 0.00024463661247864... | Human-written | 7 | Sep-24 | str | https://www.discovermagazine.com/the-sciences/hand-gestures-may-have-been-the-start-of-human-language | Human-Generated | {'text': 'Language gave humans a leg up compared to all other species. By working together outside of the family unit through spoken — and later written — communication, we were able to hunt and gather, farm, and build great civilizations.\n\nThere’s a lot we don’t know about how it all started, though. That’s because ... | {'text': 'Language gave humans a leg up compared to all other species. By working together outside of the family unit through spoken — and later written — communication, we were able to hunt and gather, farm, and build great civilizations.\n\nThere’s a lot we don’t know about how it all started, though. That’s because ... | 4 | The Sciences | Discover | jenna-russell/human_detectors | 3 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | Our unique language systems distinguish us from other species. But did early hand gestures pave the way for civilization? Researchers disagree. | Hand gestures may have sparked human language |
0.052055 | {
"comment": "The opening sentence is fun, unique, and contains some alliteration; it's exactly the kind of sentence AI wouldn't open with. The quotes sound real, especially the one about hippos spraying their faeces like a sprinkler, which is something most AI models would probably rather avoid. There are a number o... | {
"comment": "Varying sentence lengths and richer vocabulary make me think this is human-generated. Also, there are none of the typical words used by AI.",
"confidence": 4,
"guess": "Human-Generated"
} | {
"comment": "I started out thinking this was AI, and then the typos and spelling mistakes revealed it was human-generated. There is some inconsistency with capitalisation, the sentences vary in length (AI tends to stick to a strict formula) and the headings are sentence case. ",
"confidence": 5,
"guess": "Human-... | {
"comment": "I am fairly confident this is human-written, but considering the spacing pattern I've seen with previous entries, I am not 100 percent certain. To me, it's the conversational tone used throughout the article, with phrases such as \"you'd quickly realize just how different they are.\" and \"To put those ... | {
"comment": "The text is refreshingly free from excessive verbiage. The author gives me a few interesting facts without feeling the need to rabbit on about the \"impacts\" these hippopotami have on the rich tapestry of life. Formally stated: the structural features of the text are typically human.",
"confidence": ... | Moo Deng—a squishy, shiny, purple-pink pygmy hippo born this summer at a zoo in Thailand, and whose name means “bouncing pig” —has taken the internet by storm.
And that might have folks wondering, what is a pygmy hippo, anyway?
The pygmy hippopotamus (Choeropsis liberiensis) is one of two living hippo species. The ot... | Jason Bittel | Most likely human-generated | Most likely human-generated | 0.270712 | Human-Generated | gpt-4o | {'version': '2025-01-09-base', 'scanId': '32f64bdb-f59a-4fe5-b4f9-ac2ec473df53', 'documents': [{'sentences': [{'generated_prob': 0.0005146475159563124, 'sentence': 'Moo Dengᅳa squishy, shiny, purple-pink pygmy hippo born this summer at a zoo in Thailand, and whose name means “bouncing pig” ᅳhas taken the internet by st... | Human-written | 15 | 17-Sep-24 | str | https://www.nationalgeographic.com/animals/article/moo-deng-pygmy-hippo-differences | Human-Generated | {'text': 'Moo Deng—a squishy, shiny, purple-pink pygmy hippo born this summer at a zoo in Thailand, and whose name means “bouncing pig” —has taken the internet by storm.\n\nAnd that might have folks wondering, what is a pygmy hippo, anyway?\n\nThe pygmy hippopotamus (Choeropsis liberiensis) is one of two living hippo s... | {'text': 'Moo Deng—a squishy, shiny, purple-pink pygmy hippo born this summer at a zoo in Thailand, and whose name means “bouncing pig” —has taken the internet by storm.\n\nAnd that might have folks wondering, what is a pygmy hippo, anyway?\n\nThe pygmy hippopotamus (Choeropsis liberiensis) is one of two living hippo s... | 8 | Animals | National Geographic | jenna-russell/human_detectors | 7 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | With Moo Deng's adorable behavior and panicked expressions going viral, experts weigh in on this lesser-known hippo species evolved from its much larger cousins. | How Moo Deng the pygmy hippo is different from common hippos |
0.007403 | {
"comment": "A hot sauce bearing the name of a person's drag persona feels way too creative for AI to have come up with it. The same goes for a person covering their hands in Vaseline after handling chillies. The introduction pulls you in and the conclusion is somewhat humorous, pointing to human writing. AI sentenc... | {
"comment": "I'm fairly confident this text is human-generated. It uses varying lengths and sentences of different styles. AI tends to use just a few basic tenses, while this uses past perfect tense in places. Also, there are some errors with leaving out the Oxford comma. AI doesn't make this mistake. AI doesn't gen... | {
"comment": "Again their are grammatical and punctuational issues that AI wouldn't have overlooked. Rather than using a mix of tenses (for e.g.), AI would have followed a more textbook standard structure. Furthermore, I very much doubt AI would have used adventurous adjectives like 'chunky', 'musky' or 'thin' to des... | {
"comment": "While I had a harder time with this article, I believe it's human-written. Because of the topic and tone written, this article feels like one you'd find in your online local newspaper from a reporter. Sentences such as \"But the Scotch bonnet, with its distinctive, tart, numbing heat, is always at the h... | {
"comment": "Pretty sure this is written by a human. Variation of paragraph and sentence length, narrative style, and the appropriate use of normal-sounding dialogue are not usually found in AI texts. Word choice is evocative and novel analogies are used. The writer uses simple tenses and active verbs, and does not... | As a child, the cookbook author Lesley Enston learned maybe the most important lesson about pepper sauce from her uncle Richard in Trinidad and Tobago: Be very careful when handling Scotch bonnet chiles.
“I will always remember coming back to the house in Princes Town and finding him on the couch with his hands up in ... | Korsha Wilson | Most likely human-generated | Most likely human-generated | 0.715306 | Human-Generated | gpt-4o | {'version': '2025-01-09-base', 'scanId': '8c8d6cbf-7d25-46fd-b546-e579176b59b1', 'documents': [{'sentences': [{'generated_prob': 0.00011882380931638181, 'sentence': 'As a child, the cookbook author Lesley Enston learned maybe the most important lesson about pepper sauce from her uncle Richard in Trinidad and Tobago: Be... | Human-written | 25 | 9/25/24 | str | https://www.nytimes.com/2024/09/25/dining/caribbean-pepper-sauce.html | Human-Generated | {'text': 'As a child, the cookbook author Lesley Enston learned maybe the most important lesson about pepper sauce from her uncle Richard in Trinidad and Tobago: Be very careful when handling Scotch bonnet chiles.\n\n“I will always remember coming back to the house in Princes Town and finding him on the couch with his ... | {'text': 'As a child, the cookbook author Lesley Enston learned maybe the most important lesson about pepper sauce from her uncle Richard in Trinidad and Tobago: Be very careful when handling Scotch bonnet chiles.\n\n“I will always remember coming back to the house in Princes Town and finding him on the couch with his ... | 13 | Dining | New York Times | jenna-russell/human_detectors | 12 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | Across the region and the diaspora, the Scotch bonnet lends its distinctive, tart, numbing heat to this hot sauce, offering a taste of the islands and a punch of flavor. | How Caribbean Pepper Sauce Gets Its Fire |
0.274602 | {
"comment": "The way the article quotes various doctors from universities would normally make me lean towards AI, if not for it being about a medical condition. It seems human-generated, but it still has a somewhat AI style. The part about money for needy families going to the wealthy is a bit dark for AI, it also u... | {
"comment": "There are none of the usual AI-repeated words or phrases. The sentences are of varied style and length. ",
"confidence": 4,
"guess": "Human-Generated"
} | {
"comment": "I know all the tells now, which are: missing quotation marks. Speech tags. Structure. Syntax. Sentence case headings. en dashes. No Oxford comma. ",
"confidence": 5,
"guess": "Human-Generated"
} | {
"comment": "While I am less confident with this article, I feel that the base of the article is human-written. The reason I struggle with identifying this article is due to its tone and sentence structure. It uses a sentence length similar to AI generations, but structures the sentences in a highly neutral tone. Ev... | {
"comment": "The fact that all the people the author quotes only \"said\" and didn't state, voice, declare, pronounce, express, etc. makes me think that a human produced this story. I'm somewhat puzzled why the author added paragraph 2 because it has no relevance to anything else in the text and the \"scandal\" is n... | The retired N.F.L. quarterback Brett Favre told a congressional committee on Tuesday that he had been diagnosed with Parkinson’s disease, a progressive, incurable condition that causes tremors and stiffness, often resulting in impaired movement and speech.
The Republican-led committee was hearing testimony related to ... | Emily Schmall and Nina Agrawal | Most likely human-generated | Most likely human-generated | 0.621318 | Human-Generated | gpt-4o | {'version': '2025-01-09-base', 'scanId': '5282ee8a-e9b2-4df7-b542-22676becfb9c', 'documents': [{'sentences': [{'generated_prob': 1.7049862435669638e-05, 'sentence': 'The retired N.F.L.', 'perplexity': 0, 'highlight_sentence_for_ai': False}, {'generated_prob': 1.0670940355339553e-05, 'sentence': "quarterback Brett Favre... | Human-written | 27 | 9/25/24 | str | https://www.nytimes.com/article/parkinsons-disease-head-injury.html | Human-Generated | {'text': 'The retired N.F.L. quarterback Brett Favre told a congressional committee on Tuesday that he had been diagnosed with Parkinson’s disease, a progressive, incurable condition that causes tremors and stiffness, often resulting in impaired movement and speech.\n\nThe Republican-led committee was hearing testimony... | {'text': 'The retired N.F.L. quarterback Brett Favre told a congressional committee on Tuesday that he had been diagnosed with Parkinson’s disease, a progressive, incurable condition that causes tremors and stiffness, often resulting in impaired movement and speech.\n\nThe Republican-led committee was hearing testimony... | 14 | Article | New York Times | jenna-russell/human_detectors | 13 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | Brett Favre announced his diagnosis on Tuesday. Here’s what we know about the disease and its relationship to head injuries. | What Is Parkinson’s Disease? |
0.101974 | {
"comment": "The opening sentence immediately seems like it was written by a human, referencing nonalcoholic drinks as the \"butt of jokes\" just doesn't sound like something AI would ever write. The general way the subject matter is handled with references to alcoholism, damaged pregnancies, and cancer, seems a bit... | {
"comment": "I don't see any typical AI words or phrases. THere is a \"chatty\" informal tone to the writing which sounds more human than AI. The use of conjunctions at the beginning of sentences is a mistake I don't think AI usually makes.",
"confidence": 3,
"guess": "Human-Generated"
} | {
"comment": "Here's why I think it's human-generated: no Oxford comma. AI prefers to use the percentage symbol. Colloquial terms. Awkward phrasing. Missing punctuation. Text isn't very well written and sounds quite amateur. ",
"confidence": 5,
"guess": "Human-Generated"
} | {
"comment": "I am quite confident this is human-written. The article presents its information well, providing context for the topic with not only facts that are easy to understand, but also provide relevance to the quotes placed throughout it. It's a conversational tone that's relatable and reliance, as seen with ph... | {
"comment": "A considerable number of sentences begin with coordinating conjunctions, which facilitate textual coherence and flow. Two sentences use repetition as a rhetorical device, not a feature in AI text. The author uses modals that express possibility rather than certainty, whereas AI is usually pretty sure of... | Nonalcoholic beverages used to be the butt of jokes. Now, they’re the fastest-growing sector of the alcohol market. Sales of nonalcoholic beer, wine and liquor increased by 32 percent from 2022 to 2023, while total alcohol sales grew by just 1 percent. As of early 2024, the top-selling beer at Whole Foods was nonalcoho... | Dana G. Smith | Most likely human-generated | Most likely human-generated | 0.406694 | Human-Generated | gpt-4o | {'version': '2025-01-09-base', 'scanId': 'bfd386a9-4d2c-4b1a-a286-654839af90af', 'documents': [{'sentences': [{'generated_prob': 0.13552071154117584, 'sentence': 'Nonalcoholic beverages used to be the butt of jokes.', 'perplexity': 0, 'highlight_sentence_for_ai': False}, {'generated_prob': 0.11217872053384781, 'sentenc... | Human-written | 29 | 8/17/24 | str | https://www.nytimes.com/2024/08/17/well/eat/nonalcoholic-beverages-health-benefits.html | Human-Generated | {'text': 'Nonalcoholic beverages used to be the butt of jokes. Now, they’re the fastest-growing sector of the alcohol market. Sales of nonalcoholic beer, wine and liquor increased by 32 percent from 2022 to 2023, while total alcohol sales grew by just 1 percent. As of early 2024, the top-selling beer at Whole Foods was... | {'text': 'Nonalcoholic beverages used to be the butt of jokes. Now, they’re the fastest-growing sector of the alcohol market. Sales of nonalcoholic beer, wine and liquor increased by 32 percent from 2022 to 2023, while total alcohol sales grew by just 1 percent. As of early 2024, the top-selling beer at Whole Foods was... | 15 | Wellness | New York Times | jenna-russell/human_detectors | 14 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | Sales of NA beer, wine and spirits are soaring. What’s in them and how healthy are they? | What to Know About Non-Alcoholic Beverages |
0.025184 | {"comment":"The way in which everyone is introduced in the study feels very formulaic, which is typi(...TRUNCATED) | {"comment":"I'm not seeing any generic AI words and phrases. The text is rich with unusual turns of (...TRUNCATED) | {"comment":"Here's why I think it's human-generated: awkward phrasing and redundancies. Sounds sligh(...TRUNCATED) | {"comment":"This article is definitely human-written. Not only does the article present information (...TRUNCATED) | {"comment":"The structure of this text fails the AI test. There are too many discourse markers that (...TRUNCATED) | "Shortly after generative artificial intelligence hit the mainstream, researchers warned that chatbo(...TRUNCATED) | Teddy Rosenbluth | Most likely human-generated | Most likely human-generated | 0.452762 | Human-Generated | gpt-4o | "{'version': '2025-01-09-base', 'scanId': 'a94b86d7-5db1-4236-a261-c87a2f12d1fe', 'documents': [{'se(...TRUNCATED) | Human-written | 31 | 9/12/24 | str | https://www.nytimes.com/2024/09/12/health/chatbot-debunk-conspiracy-theories.html | Human-Generated | "{'text': 'Shortly after generative artificial intelligence hit the mainstream, researchers warned t(...TRUNCATED) | "{'text': 'Shortly after generative artificial intelligence hit the mainstream, researchers warned t(...TRUNCATED) | 16 | Health | New York Times | jenna-russell/human_detectors | 15 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | "In a new study, many people doubted or abandoned false beliefs after a short conversation with the (...TRUNCATED) | This Chatbot Pulls People Away From Conspiracy Theories |
0.00689 | {"comment":"Referring to a patch of Escherichia coli (which I'm guessing is E. coli) as \"a tasty sn(...TRUNCATED) | {"comment":"There are none of the words AI usually favors. There are a few mistakes AI probably woul(...TRUNCATED) | {"comment":"This one was difficult. I couldn't see anything glaringly wrong with it, but the missing(...TRUNCATED) | {"comment":"This article feels human-written. I believe it's human-written because while there is a (...TRUNCATED) | {"comment":"Some of the author's assertions are so garbled that only a human who doesn't quite under(...TRUNCATED) | "Scientists have given artificial intelligence a direct line into the nervous systems of millimeter-(...TRUNCATED) | Matthew Hutson | Most likely human-generated | Most likely human-generated | 0.471249 | Human-Generated | gpt-4o | "{'version': '2025-01-09-base', 'scanId': '3bb2beec-92f9-42f7-835c-2bc21aac306a', 'documents': [{'se(...TRUNCATED) | Human-written | 39 | 9/2/24 | str | https://www.scientificamerican.com/article/scientists-make-cyborg-worms-with-a-brain-guided-by-ai/ | Human-Generated | "{'text': 'Scientists have given artificial intelligence a direct line into the nervous systems of m(...TRUNCATED) | "{'text': 'Scientists have given artificial intelligence a direct line into the nervous systems of m(...TRUNCATED) | 20 | Artificial Intelligence | Scientific American | jenna-russell/human_detectors | 19 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | AI and tiny worms team up to get to treats | Scientists Make ‘Cyborg Worms’ with a Brain Guided by AI |
0.004172 | {"comment":"The writer's general attitude towards Halloween's trick or treating \"begging for treats(...TRUNCATED) | {"comment":"There's a richness of language in this text that makes me think it's human-generated. Th(...TRUNCATED) | {"comment":"I could tell from the first sentence that this was written by a human. It feels more col(...TRUNCATED) | {"comment":"AI models, in my opinion, struggles with word choice and flow, and in this article, I fe(...TRUNCATED) | {"comment":"The text is informative without being overwhelming or tedious. Judicious use of punctuat(...TRUNCATED) | "Between themed TV specials, novelty candy flavors and the pageantry of the mass-produced costume, H(...TRUNCATED) | Brandon Tensley | Most likely human-generated | Most likely human-generated | 0.528075 | Human-Generated | gpt-4o | "{'version': '2025-01-09-base', 'scanId': 'bbba742b-c687-4cc0-aa8a-4969d3d33a30', 'documents': [{'se(...TRUNCATED) | Human-written | 45 | September/October 2024 | str | https://www.smithsonianmag.com/history/celtic-origins-trick-treating-halloween-180984916/ | Human-Generated | "{'text': 'Between themed TV specials, novelty candy flavors and the pageantry of the mass-produced (...TRUNCATED) | "{'text': 'Between themed TV specials, novelty candy flavors and the pageantry of the mass-produced (...TRUNCATED) | 23 | History | Smithsonian Magazine | jenna-russell/human_detectors | 22 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | The spine-tingling roots of a mischievous Halloween tradition | The Celtic Origins of Trick-or-Treating |
0.005875 | {"comment":"The author has a unique writing style, which can most easily be seen in their use of \"s(...TRUNCATED) | {"comment":"The text has a nuanced style and some personal anecdotes, suggesting it was likely writt(...TRUNCATED) | {"comment":"I'm going to say human-generated for this one as there definitely should not be a questi(...TRUNCATED) | {"comment":"Funny enough, while AI likes to use the words \"journey\" or \"embark\" as part of its l(...TRUNCATED) | {"comment":"Diversity in sentence and paragraph length. The author references himself as \"I\" and t(...TRUNCATED) | "In February 1973, Dave Arneson, a history major at the University of Minnesota and part-time securi(...TRUNCATED) | David M. Perry | Most likely human-generated | Most likely human-generated | 0.526766 | Human-Generated | gpt-4o | "{'version': '2025-01-09-base', 'scanId': 'db712273-dbb4-490b-861a-25b1af149eb0', 'documents': [{'se(...TRUNCATED) | Human-written | 49 | July/August 2024 | str | https://www.smithsonianmag.com/arts-culture/how-dungeons-dragons-sparked-revolution-how-we-play-everything-180984498/ | Human-Generated | "{'text': 'In February 1973, Dave Arneson, a history major at the University of Minnesota and part-t(...TRUNCATED) | "{'text': 'In February 1973, Dave Arneson, a history major at the University of Minnesota and part-t(...TRUNCATED) | 25 | Arts & Culture | Smithsonian Magazine | jenna-russell/human_detectors | 24 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | "Created more than 50 years ago, the game has captured the imaginations of generations of Americans,(...TRUNCATED) | How Dungeons & Dragons Sparked a Revolution in How We Play Just About Everything |
0.091028 | {"comment":"The atmosphere of the article is overall much too dark for AI. There's death, war, menta(...TRUNCATED) | {"comment":"There's broader vocabulary than used by AI such as in the first highlighted sentence (\"(...TRUNCATED) | {"comment":"Some of the phrasing sounds very human. For example, AI would say something like 'receiv(...TRUNCATED) | {"comment":"I'll admit, I am less confident with this one than I was with the other ones, because in(...TRUNCATED) | {"comment":"The opening paragraph has one sentence. Lots of proper nouns, references to time, and pl(...TRUNCATED) | "On the day Asae Hiyama gave birth to a son, her husband left to serve in the Imperial Japanese Army(...TRUNCATED) | Ellen Wexler | Most likely human-generated | Most likely human-generated | 0.298678 | Human-Generated | gpt-4o | "{'version': '2025-01-09-base', 'scanId': '381fb7f7-8018-45e1-80c5-3e8a3e9f0328', 'documents': [{'se(...TRUNCATED) | Human-written | 51 | 9/23/24 | str | https://www.smithsonianmag.com/smart-news/a-japanese-soldiers-son-receives-a-memento-of-his-father-who-was-killed-during-world-war-ii-180985121/ | Human-Generated | "{'text': 'On the day Asae Hiyama gave birth to a son, her husband left to serve in the Imperial Jap(...TRUNCATED) | "{'text': 'On the day Asae Hiyama gave birth to a son, her husband left to serve in the Imperial Jap(...TRUNCATED) | 26 | Smart News | Smithsonian Magazine | jenna-russell/human_detectors | 25 | 1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a | https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json | validation | prompt_id_grouped_sha256_seed | tell-human-detectors-v1-2242 | "The so-called good-luck flag, which hung on an American veteran’s wall for many years, returned h(...TRUNCATED) | A Japanese Soldier’s Son Receives a Memento of His Father, Who Was Killed During World War II |
End of preview. Expand in Data Studio
TELL Human Detectors Split
This dataset is a prompt-disjoint validation/test split of human_detectors.json from
Jenna Russell, Marzena Karpinska, and Mohit Iyyer, "People who frequently use
ChatGPT for writing tasks are accurate and robust detectors of AI-generated text"
(arXiv:2501.15654).
The split is intended as a research artifact for evaluating AI-vs-human writing
detectors and explanation-quality methods. It preserves the upstream fields, with
one documented issue type normalization for Hugging Face compatibility, and adds
split provenance metadata.
This TELL split artifact was prepared by Suraj Ranganath and Aldan Creo. The underlying Human Detectors dataset should be cited to Russell, Karpinska, and Iyyer.
Source
- Upstream repository: https://github.com/jenna-russell/human_detectors
- Source file:
https://raw.githubusercontent.com/jenna-russell/human_detectors/refs/heads/main/human_detectors.json - Source dataset license: MIT
- Canonical source SHA256:
1b8f9eeb6413f2541a320b09c7cc77b8d5c51a9e69a87ae0e91bd0e9ede5bd9a
Split Protocol
The original dataset has 300 documents, 30 prompt_id groups, and 10 rows per
prompt group. Each prompt group contains exactly one row for each combination of
ground_truth in AI-generated, Human-written and generation_model in
claude, gpt-4o, humanized_o1-pro, o1-pro, and paraphrased_gpt-4o.
We split by prompt_id, not by individual row, to prevent the same underlying
prompt/topic from appearing in both validation and test.
- Split seed:
tell-human-detectors-v1-2242 - Validation prompt IDs:
[4, 8, 13, 14, 15, 16, 20, 23, 25, 26] - Test prompt IDs:
[1, 2, 3, 5, 6, 7, 9, 10, 11, 12, 17, 18, 19, 21, 22, 24, 27, 28, 29, 30] - Validation rows: 100
- Test rows: 200
Both splits are balanced by label and generation model. Validation has 50 human-written and 50 AI-generated documents. Test has 100 human-written and 100 AI-generated documents.
Fields
All upstream fields are preserved, including article metadata, article,
ground_truth, automatic detector outputs, five annotator objects, majority_vote,
and expert_majority_vote. The upstream issue field mixes strings and integers,
so this artifact stores issue as a string for Hugging Face/Arrow compatibility
and records the original Python type in issue_original_type.
Additional fields:
split:validationortestsource_row_index: integer row index in the sorted upstream JSON objectsource_dataset: upstream dataset identifiersource_url: upstream source URLsource_sha256: canonical upstream checksum used by this artifactsplit_method:prompt_id_grouped_sha256_seedsplit_seed: deterministic split seedissue_original_type: original upstream JSON type forissue
Recommended Use
Use validation for prompt, metric, and model-selection decisions. Use test once
for final reporting. Do not tune on the test set, and do not merge validation and
test without reporting that choice.
Citation
@misc{russell2025humandetectors,
title={People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text},
author={Russell, Jenna and Karpinska, Marzena and Iyyer, Mohit},
year={2025},
eprint={2501.15654},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
- Downloads last month
- 112