Datasets:

G-reen
/

fastdetector-test-stat-test-mini

original string	prompt dict	topic string	format string	response_0 string	response_1 string	generator_model string	generation_params string	final_response string	jaccard_1 float64	jaccard_2 float64	levenshtein float64	softngram float64	cosdist float64	bertscore_precision float64	bertscore_recall float64	bertscore float64	moverscore float64	reranker float64	original_perplexity_llama_instruct float64	original_entropy_llama_instruct float64	original_topp_outlier_llama_instruct float64	original_topk_outlier_llama_instruct float64	original_fastdetectgpt_llama_instruct float64	original_perplexity_llama_base float64	original_entropy_llama_base float64	original_topp_outlier_llama_base float64	original_topk_outlier_llama_base float64	original_fastdetectgpt_llama_base float64	final_response_perplexity_llama_instruct float64	final_response_entropy_llama_instruct float64	final_response_topp_outlier_llama_instruct float64	final_response_topk_outlier_llama_instruct float64	final_response_fastdetectgpt_llama_instruct float64	final_response_perplexity_llama_base float64	final_response_entropy_llama_base float64	final_response_topp_outlier_llama_base float64	final_response_topk_outlier_llama_base float64	final_response_fastdetectgpt_llama_base float64	original_binoculars float64	final_response_binoculars float64	original_editlens_bucket_roberta_large int64	original_editlens_score_roberta_large float64	final_response_editlens_bucket_roberta_large int64	final_response_editlens_score_roberta_large float64	original_editlens_score_llama_3_2_3b float64	final_response_editlens_bucket_llama_3_2_3b int64	final_response_editlens_score_llama_3_2_3b float64	original_editlens_score_giga_roberta float64	final_response_editlens_bucket_giga_roberta int64	final_response_editlens_score_giga_roberta float64	original_editlens_score_giga_llama_3_2_3b float64	final_response_editlens_bucket_giga_llama_3_2_3b int64	final_response_editlens_score_giga_llama_3_2_3b float64	original_llmjudge_valid_qwen38_27b int64	final_response_llmjudge_score_qwen38_27b float64	final_response_llmjudge_valid_qwen38_27b int64	original_llmjudge_valid_gemma4_31b int64	final_response_llmjudge_score_gemma4_31b float64	final_response_llmjudge_valid_gemma4_31b int64
2016 has been an eventful year for the video game industry, to say the least, and it’s safe to say that a lot of these aforementioned events have centered around controversy. Whether it be regarding alleged leaks for anticipated hardware, or massive backlash against hugely anticipated releases, there’s just been a lot ...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nWrite a different piece of text on the same topic.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have written.\nThe final text must have the...	Games	Listicle	The games industry rarely has a quiet year, but 2016 seemed especially determined to turn every announcement, delay, and design decision into a major controversy. Players had legitimate reasons to criticize publishers and developers, yet some debates quickly escalated far beyond the importance of the original issue. Fi...		gpt-5.6-sol	{"reasoning_effort": "none"}	The games industry rarely has a quiet year, but 2016 seemed especially determined to turn every announcement, delay, and design decision into a major controversy. Players had legitimate reasons to criticize publishers and developers, yet some debates quickly escalated far beyond the importance of the original issue. Fi...	0.835084	0.969546	5,448	0.949796	0.16365	0.184048	0.195169	0.189647	0.666543	-9.1875	11.097288	2.370328	0.049748	0.078086	-0.688457	9.007109	2.193631	0.034005	0.06927	-0.088258	18.820877	2.950154	0.034989	0.115124	0.199514	14.641636	2.711716	0.044018	0.092551	0.384813	0.823254	0.813209	0	0.029816	3	0.964348	0.052568	3	0.760888	0.003012	3	0.998568	0.003214	3	0.670748	8	0.9375	8	8	1	8
Some 650 years old, an illuminated Haggadah - the text read by pious Jews annually to commemorate the Passover, survives in Bosnia-Herzogovina, having perhaps miraculously survived centuries of anti-Semitic persecution. Now known as the Sarajevo Haggadah, it has survived Spain's expulsion of the Jews following the Alha...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nEdit this text by replacing five to ten percent of words in it, without changing anything else.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what yo...	History	Nonfiction Writing	Some 650 years old, an illuminated Haggadah - the text recited by devout Jews annually to observe the Passover, survives in Bosnia-Herzegovina, having almost miraculously endured centuries of anti-Semitic persecution. Now called the Sarajevo Haggadah, it has survived Spain's expulsion of the Jews after the Alhambra Dec...	Some 650 years old, an illuminated Haggadah - the text recited by devout Jews annually to observe the Passover, survives in Bosnia-Herzegovina, having almost miraculously endured centuries of anti-Semitic persecution. Now called the Sarajevo Haggadah, it has survived Spain's expulsion of the Jews after the Alhambra Dec...	gpt-5.6-sol	{"reasoning_effort": "none"}	Some 650 years old, an illuminated Haggadah - the text recited by devout Jews annually to observe the Passover, survives in Bosnia-Herzegovina, having almost miraculously endured centuries of anti-Semitic persecution. Now called the Sarajevo Haggadah, it has survived Spain's expulsion of the Jews after the Alhambra Dec...	0.396825	0.515556	561	0.060137	0.004722	0.038948	0.038553	0.038751	0.251601	-9.375	10.888336	2.280931	0.039788	0.091512	-1.348529	8.769365	2.191471	0.037135	0.068966	0.269972	12.039354	2.329077	0.049598	0.091153	-2.012751	9.768044	2.209138	0.038874	0.075067	-0.937493	0.860734	0.884923	0	0.029024	0	0.04356	0.005783	0	0.031172	0.003108	0	0.008683	0.002931	0	0.036293	8	0	8	8	0.125	8
The long-awaited trial by the International Tribunal at the Hague of Radovan Karadzic, one of the Serbian leaders responsible for the Bosnian genocide and accused of crimes against humanity has not yet taken off. Like his mentor Milosevic who actually died before sentencing—a death that in its timeliness, one might thi...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nTake a subpoint within this text and write a full new text about it, using the language and wording from the text as appropriate.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a ...	Entertainment	Nonfiction Writing	The most gripping conflict in Storm is not simply between prosecutor Hannah Maynard and the Serb war criminal she is trying to convict. It is between ethics and expediency, played out inside the bureaucracy of international organizations whose officials are meant to serve justice but are often more interested in making...		gpt-5.6-sol	{"reasoning_effort": "none"}	The most gripping conflict in Storm is not simply between prosecutor Hannah Maynard and the Serb war criminal she is trying to convict. It is between ethics and expediency, played out inside the bureaucracy of international organizations whose officials are meant to serve justice but are often more interested in making...	0.722857	0.90106	1,553	0.805911	0.207483	0.151745	0.173053	0.162534	0.612586	-8.3125	24.575876	3.095208	0.04	0.162353	-0.941159	19.349444	2.815809	0.04	0.134118	-1.351124	21.476192	2.905817	0.039326	0.140449	-1.353657	17.565334	2.941546	0.030899	0.123596	0.643532	0.870203	0.897228	0	0.141591	3	0.994066	0.039154	1	0.510405	0.002762	3	0.999867	0.003736	0	0.17673	8	0.125	8	8	0.375	8
At 3-6, the Jacksonville Sharks have run out of mulligans. Winning the last nine regular-season games may not be enough to win the AFL American Conference South Division title for the fifth consecutive year. But any more losses would move the Sharks from a difficult to a practically impossible situation. “We are in a h...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nWrite a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text.\nOutput the full new text with no extra statements or commentations.\nB...	Sports & Fitness	News Article	At 3-6, the Jacksonville Sharks are in trouble, but saying they have “run out of mulligans” overstates the finality of their situation. With nine regular-season games remaining, Jacksonville still has half a season to correct its mistakes. Winning the AFL American Conference South Division for a fifth consecutive year...		gpt-5.6-sol	{"reasoning_effort": "none"}	At 3-6, the Jacksonville Sharks are in trouble, but saying they have “run out of mulligans” overstates the finality of their situation. With nine regular-season games remaining, Jacksonville still has half a season to correct its mistakes. Winning the AFL American Conference South Division for a fifth consecutive year ...	0.731092	0.898266	2,630	0.876002	0.078358	0.15832	0.132809	0.145755	0.593845	-8.25	9.684573	2.072132	0.044266	0.054326	-2.247109	6.79623	2.065039	0.036217	0.038229	1.780785	13.71777	2.350283	0.059783	0.088315	-3.515664	10.491061	2.290331	0.048913	0.067935	-0.817345	0.775385	0.862557	0	0.092957	0	0.221842	0.031015	1	0.422505	0.006216	0	0.037083	0.005437	1	0.325324	8	0.4375	8	8	0.75	8
2 breakout games in a row for the rookie... his playing style reminds me alot of Pau when he first came into the league... his european playing style compliments Yao... and if he can play better defence and hustle for more boards, he might very well be the final piece to Houston's playoff run... this guy is a gem and h...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nCreate the shortest possible prompt that can be sent to an LLM to reproduce this text. The prompt may include verbatim pieces of the original text, however, it must be short and accurate, working as an effective compression of the original text.\nDo not outpu...	Sports & Fitness	Comment Section	Reproduce this 5-post Rockets forum thread verbatim: rookie Scola’s two breakout games/Pau comparison/complements Yao/needs defense and boards; don’t overhype, foul trouble, Chuck Hayes enables Yao doubles while Scola’s offense earns respect; PF now set but Rafer+James+Francis=Larry Hughes; ask why Rockets—not Wizards—...	Comment Section User1: After these two breakout games, Scola is looking like a rookie version of Pau Gasol. His offensive game complements Yao perfectly. If he can improve his defense and rebounding, we’ve got something special. User2: Let’s not overhype him after two games. He still gets into foul trouble. Chuck Hay...	gpt-5.6-sol	{"reasoning_effort": "none"}	Comment Section User1: After these two breakout games, Scola is looking like a rookie version of Pau Gasol. His offensive game complements Yao perfectly. If he can improve his defense and rebounding, we’ve got something special. User2: Let’s not overhype him after two games. He still gets into foul trouble. Chuck Hayes...	0.835206	0.963824	970	0.845503	0.129362	0.176505	0.173856	0.175183	0.635726	-7.5625	39.474282	3.195621	0.071429	0.180124	-3.442785	24.738754	3.085403	0.046584	0.127329	-0.926837	30.380973	2.953886	0.074866	0.197861	-2.674152	21.159692	3.000661	0.053476	0.15508	-0.301555	0.8551	0.875319	0	0.019062	1	0.443046	0.018798	1	0.311482	0.006563	2	0.554442	0.006273	1	0.277346	8	0	8	8	1	8
Welcome Parents and Family Members. Disney College Program is a paid internship offering participants the opportunity to advance their strengths and interests, meet guests and cast members from around the world and take part in educational opportunities students can't get anywhere else. This truly unique program allows...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nEdit this text by removing all adverbs, making only minimal adjustments to the surrounding grammar to ensure the sentences remain structurally sound.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do...	Education & Jobs	Customer Support	Welcome Parents and Family Members. Disney College Program is a paid internship offering participants the opportunity to advance their strengths and interests, meet guests and cast members from around the world and take part in educational opportunities students can't get elsewhere. This program allows participants to ...	Welcome Parents and Family Members. Disney College Program is a paid internship offering participants the opportunity to advance their strengths and interests, meet guests and cast members from around the world and take part in educational opportunities students can't get elsewhere. This program allows participants to ...	gpt-5.6-sol	{"reasoning_effort": "none"}	Welcome Parents and Family Members. Disney College Program is a paid internship offering participants the opportunity to advance their strengths and interests, meet guests and cast members from around the world and take part in educational opportunities students can't get elsewhere. This program allows participants to ...	0.625	0.682759	1,409	0.332106	0.282017	0.071562	0.206877	0.144538	0.609048	-1.9375	15.23184	2.524051	0.060475	0.090713	-1.957022	12.435065	2.388437	0.056156	0.086393	-1.410896	15.329064	2.561653	0.066079	0.101322	-1.17732	13.116509	2.484454	0.044053	0.110132	-0.677065	0.90381	0.903486	0	0.008239	0	0.01418	0.00165	0	0.034436	0.001448	0	0.002919	0.001431	0	0.008144	8	0.5625	8	8	0.5625	8
You state: "the primary myth that I am trying to debunk in the article is that either Cocoa or Carbon are the one true API for OS X." A, potentially, laudable goal, but not, entirely, true---at least from an historic perspective (thus, it's subject to change as things evolve). When you look at the history of this syste...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nWrite a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text.\nOutput the full new text with no extra statements or commentations.\nB...	Software Dev.	Comment Section	The historical argument is less decisive than it appears, because chronology does not establish architectural primacy, technical superiority, or long-term strategic intent. That BSD, Mach, and the OpenStep-derived frameworks existed before Carbon does not make Cocoa the uniquely authentic API for Mac OS X. Mac OS X was...		gpt-5.6-sol	{"reasoning_effort": "none"}	The historical argument is less decisive than it appears, because chronology does not establish architectural primacy, technical superiority, or long-term strategic intent. That BSD, Mach, and the OpenStep-derived frameworks existed before Carbon does not make Cocoa the uniquely authentic API for Mac OS X. Mac OS X was...	0.900504	0.983905	7,035	0.95371	0.145473	0.224225	0.178812	0.202164	0.699037	-8.6875	31.060908	3.369589	0.047404	0.173815	-0.550962	22.75617	3.082358	0.040632	0.121896	-0.382763	27.291693	3.041763	0.048932	0.166782	-4.246779	19.429333	2.937145	0.038594	0.12612	-0.50131	0.835056	0.881135	0	0.034401	1	0.32748	0.023056	3	0.791815	0.005165	3	0.999522	0.005259	1	0.512121	8	0.0625	8	8	0.875	8
Ever hear of this kid named Justin Bieber? We’re guessing you have—and if not, chances are you’re just a few light-years out of the loop. The pint-sized pop star has caused a mall riot, has teenage girls swooning at the mere mention of his name, and saw his recently-released My World 2.0 album hit #1 and sell 283,000 c...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nTake inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should make sure there are slight, subtle grammatical mistakes, to make it sound like it was written by a human.\nOutput...	Entertainment	Audio Transcript	Ever hear of this kid named Mateo Cruz? We’re guessing you have—and if not, chances are your algorithm been sleeping. The 18-year-old singer-producer turned a bedroom-made track into a streaming monster, had thousands of fans packing a downtown skate park, and watched his debut EP, After Midnight, hit the top five less...		gpt-5.6-sol	{"reasoning_effort": "none"}	Ever hear of this kid named Mateo Cruz? We’re guessing you have—and if not, chances are your algorithm been sleeping. The 18-year-old singer-producer turned a bedroom-made track into a streaming monster, had thousands of fans packing a downtown skate park, and watched his debut EP, After Midnight, hit the top five less...	0.800587	0.952998	2,989	0.982888	0.545323	0.173676	0.175375	0.174526	0.627936	-2.125	11.847598	2.182532	0.067857	0.085714	-4.075579	8.541627	2.171834	0.05119	0.054762	0.394777	12.506175	2.13626	0.06269	0.091001	-6.112204	10.533939	2.220089	0.048534	0.073812	-2.152618	0.845998	0.950104	1	0.279887	1	0.522639	0.04863	1	0.374033	0.011073	1	0.265918	0.003414	0	0.134972	8	0.3125	8	8	1	8
While there are still plenty of collective thought posts I can write, today I decided to just share a few of my random thoughts about the trip, as well as some random photos. And for some reason, it only wanted me to include these photos at the top of the blog, so, that is why it is laid out this way today. These are t...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nWrite a different piece of text on the same topic.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have written.\nThe final text must have the...	Travel	Personal Blog	Some travel memories become full stories, while others remain small moments that are just as meaningful. Looking back on my trip, I keep thinking about the unexpected details—the things that surprised me, made me laugh, or helped me see everyday life differently. In Brisas del Mar, people were curious about my tattoos...		gpt-5.6-sol	{"reasoning_effort": "none"}	Some travel memories become full stories, while others remain small moments that are just as meaningful. Looking back on my trip, I keep thinking about the unexpected details—the things that surprised me, made me laugh, or helped me see everyday life differently. In Brisas del Mar, people were curious about my tattoos ...	0.75375	0.926322	4,061	0.791682	0.09014	0.137758	0.140544	0.139153	0.56377	-5.3125	15.585222	2.6242	0.062274	0.086894	-2.108185	12.574488	2.48521	0.057205	0.071687	-0.842817	13.024234	2.196963	0.071354	0.089192	-5.923295	10.696381	2.351078	0.035677	0.071354	-0.296349	0.869297	0.943368	0	0.010999	1	0.544529	0.06883	2	0.670701	0.002749	3	0.796746	0.005438	2	0.56956	8	0.125	8	8	0.75	8
Gender Gender is about how societies see men and women in different contexts, leading to them, in some cases, being valued differently. It is well known in most societies that women were, historically, considered of less importance than men - unable to carry out certain roles and responsibilities, often subservient or ...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nWrite a followup to this piece of text, in the same style and language as the text.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have writt...	Politics	About (Org.)	About (Org.) and topic Politics. Politics is about how societies make decisions, distribute resources and represent the needs of their people. It is through political systems that laws are created, services are funded and communities are given opportunities to shape their own futures. Yet, in many societies throughout...		gpt-5.6-sol	{"reasoning_effort": "none"}	About (Org.) and topic Politics. Politics is about how societies make decisions, distribute resources and represent the needs of their people. It is through political systems that laws are created, services are funded and communities are given opportunities to shape their own futures. Yet, in many societies throughout ...	0.724138	0.872117	1,213	0.908803	0.603856	0.191254	0.167841	0.179714	0.630234	0.25	21.411559	2.989384	0.06051	0.11465	-0.573089	16.035936	2.69238	0.06051	0.082803	-0.680358	12.807799	2.308947	0.052308	0.061538	-2.129042	10.450517	2.425114	0.030769	0.043077	0.714011	0.83589	0.892207	0	0.009553	0	0.148356	0.002934	1	0.541396	0.002014	0	0.027524	0.001066	0	0.011332	8	0.8125	8	8	1	8
To win a Red Dot design award, a product needs more than just a good look. The Red Dot award is about form, but more importantly – function. Ventura isn’t a stranger to either of them. In 2012, the Ventura SPARC wrist watch received a Red Dot award. The European awards outfit skews towards modern devices, and the SPARC...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nChange the tone of this text to be more conversational and approachable.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have written." ], ...	Hardware	Nonfiction Writing	Winning a Red Dot Design Award takes more than simply looking good. The award recognizes form, of course, but function matters even more—and Ventura knows a thing or two about both. In 2012, the Ventura SPARC wristwatch received a Red Dot Award. Since the European competition tends to favor modern, innovative devices, ...		gpt-5.6-sol	{"reasoning_effort": "none"}	Winning a Red Dot Design Award takes more than simply looking good. The award recognizes form, of course, but function matters even more—and Ventura knows a thing or two about both. In 2012, the Ventura SPARC wristwatch received a Red Dot Award. Since the European competition tends to favor modern, innovative devices, ...	0.641949	0.836594	1,350	0.317829	0.013011	0.092955	0.091784	0.09237	0.434113	-8.375	22.86076	3.24134	0.051908	0.141985	1.197691	17.823002	2.842295	0.047328	0.122137	-0.439333	14.408541	2.488576	0.045872	0.100917	-2.150576	11.933276	2.4469	0.030581	0.085627	-0.401644	0.810529	0.903006	0	0.01258	1	0.384377	0.012987	1	0.287408	0.001948	0	0.036058	0.001377	0	0.062084	8	0	8	8	0	8
The University Union strives to reduce the risk of crucial data loss. The Disaster Recovery Policy recognizes that, while risks of loss are always present, there are steps that can be taken to manage them responsibly. Guidelines and Procedures The University Union Operation of CSU, Sacramento will maintain a secure and...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nChange the tone of this text to be more direct and bold.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have written.\nThe final text must ha...	Software	Legal Notices	The University Union must protect critical software and data from loss. Although no system can eliminate every risk, this Disaster Recovery Policy establishes clear requirements for securing data, maintaining reliable backups, and restoring essential software services quickly. The University Union Operation of CSU, Sa...		gpt-5.6-sol	{"reasoning_effort": "none"}	The University Union must protect critical software and data from loss. Although no system can eliminate every risk, this Disaster Recovery Policy establishes clear requirements for securing data, maintaining reliable backups, and restoring essential software services quickly. The University Union Operation of CSU, Sac...	0.546992	0.773768	2,561	0.382428	0.037894	0.097311	0.106577	0.101968	0.467716	-8.5	13.240544	2.459586	0.054196	0.098776	-1.919871	11.138783	2.302968	0.051573	0.085664	-1.770391	14.496043	2.413184	0.075875	0.10214	-3.845297	11.58568	2.285777	0.051556	0.082685	-2.584001	0.903777	0.9263	0	0.013885	0	0.176439	0.025205	0	0.023118	0.001999	0	0.04181	0.002745	0	0.034828	8	0	8	8	0	8
Scotland’s wildlife is unbeatable. The Cairngorms are central to the BBC Autumn and Spring watch programmes and the wildlife listed are within or near the Cairngorms National Park. With an ongoing Wildlife Diary, Scot Mountain Holidays can direct you to where you can witness these stunning creatures. The Moray Firth Wi...	{ "chat_turns": [ "<document>\n{{DOC}}\n</document>\n\nEdit this text to eliminate redundant and filler words.\nOutput the full new text with no extra statements or commentations.\nBegin directly with the text itself. Do not add a title, a heading, or a label naming what you have written.\nThe final text must hav...	Travel	Knowledge Article	Scotland offers exceptional wildlife viewing. The Cairngorms feature prominently in the BBC’s Springwatch and Autumnwatch, and the species below can be found within or near Cairngorms National Park. Scot Mountain Holidays maintains a wildlife diary and can recommend viewing locations. The Moray Firth Wildlife Cent...		gpt-5.6-sol	{"reasoning_effort": "none"}	Scotland offers exceptional wildlife viewing. The Cairngorms feature prominently in the BBC’s Springwatch and Autumnwatch, and the species below can be found within or near Cairngorms National Park. Scot Mountain Holidays maintains a wildlife diary and can recommend viewing locations. The Moray Firth Wildlife Centr...	0.65586	0.845619	1,623	0.337438	0.036077	0.104594	0.118752	0.11173	0.503776	-7.5625	17.17196	2.689964	0.051471	0.111765	-1.781843	13.732558	2.532783	0.036765	0.098529	-1.075126	11.385176	2.282934	0.038153	0.088353	-1.599949	9.502904	2.2927	0.034137	0.084337	0.456205	0.877125	0.868593	0	0.012202	0	0.143766	0.021565	1	0.252264	0.00562	1	0.412037	0.003882	1	0.321405	8	0	8	8	0	8

End of preview. Expand in Data Studio

Auto-generated FastDetector dataset

Fastdetector Test Stat Test Mini

Best detector

Giga EditLens Llama-3.2-3B Score

0.5895 TPR @ 1% FPR

Hardest prompt subset

rewrite

0.1951 max detector TPR @ 1% FPR

Hardest generator config

claude-opus-5 (Temp: Unknown)

0.3333 max detector TPR @ 1% FPR

1,569

rows

generator configs

prompt subsets

detectors

01Leaderboard 02Model analytics 03Distances 04Appendix

Detector leaderboard

Score-based detectors ranked by overall AUROC. Thresholds are placed exactly on every human score, so TPR is reported at a true 1% and 0.1% false positive rate. Open a row for its threshold sweep, its TPR as AI rows with low cosdist are dropped, and its score distributions.

01Giga EditLens Llama-3.2-3B ScoreN3,138AUROC0.8996TPR @ 1% FPR0.5895TPR @ 0.1% FPR0.4978⌄

1% FPR threshold 0.07110.1% FPR threshold 0.1835

Threshold sweep: Giga EditLens Llama-3.2-3B Score

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Giga EditLens Llama-3.2-3B Score: TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Giga EditLens Llama-3.2-3B Score: overall human

Overall · human onlyOpen full size ↗

Giga EditLens Llama-3.2-3B Score: overall AI

Overall · AI onlyOpen full size ↗

Giga EditLens Llama-3.2-3B Score: prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Giga EditLens Llama-3.2-3B Score: generator subsets

AI texts · one envelope per generator configOpen full size ↗

02EditLens Llama-3.2-3B ScoreN3,138AUROC0.8802TPR @ 1% FPR0.5825TPR @ 0.1% FPR0.4863⌄

1% FPR threshold 0.18450.1% FPR threshold 0.2998

Threshold sweep: EditLens Llama-3.2-3B Score

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

EditLens Llama-3.2-3B Score: TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

EditLens Llama-3.2-3B Score: overall human

Overall · human onlyOpen full size ↗

Overall · AI onlyOpen full size ↗

EditLens Llama-3.2-3B Score: prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

EditLens Llama-3.2-3B Score: generator subsets

AI texts · one envelope per generator configOpen full size ↗

03Giga EditLens Roberta ScoreN3,138AUROC0.8667TPR @ 1% FPR0.5545TPR @ 0.1% FPR0.3002⌄

1% FPR threshold 0.12390.1% FPR threshold 0.6720

Threshold sweep: Giga EditLens Roberta Score

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Giga EditLens Roberta Score: TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Giga EditLens Roberta Score: overall human

Overall · human onlyOpen full size ↗

Overall · AI onlyOpen full size ↗

Giga EditLens Roberta Score: prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Giga EditLens Roberta Score: generator subsets

AI texts · one envelope per generator configOpen full size ↗

04EditLens Roberta-Large ScoreN3,138AUROC0.8550TPR @ 1% FPR0.4270TPR @ 0.1% FPR0.1842⌄

1% FPR threshold 0.31330.1% FPR threshold 0.7948

Threshold sweep: EditLens Roberta-Large Score

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

EditLens Roberta-Large Score: TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

EditLens Roberta-Large Score: overall human

Overall · human onlyOpen full size ↗

EditLens Roberta-Large Score: overall AI

Overall · AI onlyOpen full size ↗

EditLens Roberta-Large Score: prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

EditLens Roberta-Large Score: generator subsets

AI texts · one envelope per generator configOpen full size ↗

05LLM judge: Gemma-4-31B-it (thinking on, 8 samples)N3,138AUROC0.7431TPR @ 1% FPR0.3435TPR @ 0.1% FPR0.0000⌄

1% FPR threshold 0.87500.1% FPR threshold 1.0000

Threshold sweep: LLM judge: Gemma-4-31B-it (thinking on, 8 samples)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): overall human

Overall · human onlyOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): overall AI

Overall · AI onlyOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): generator subsets

AI texts · one envelope per generator configOpen full size ↗

06LLM judge: Qwen3.8-27B (reasoning medium, 8 samples)N3,138AUROC0.7179TPR @ 1% FPR0.2811TPR @ 0.1% FPR0.1383⌄

1% FPR threshold 0.50000.1% FPR threshold 0.9375

Threshold sweep: LLM judge: Qwen3.8-27B (reasoning medium, 8 samples)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): overall human

Overall · human onlyOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): overall AI

Overall · AI onlyOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): generator subsets

AI texts · one envelope per generator configOpen full size ↗

07BinocularsN3,138AUROC0.5933TPR @ 1% FPR0.0771TPR @ 0.1% FPR0.0306⌄

1% FPR threshold 0.94670.1% FPR threshold 0.9818

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Overall · human onlyOpen full size ↗

Overall · AI onlyOpen full size ↗

AI texts · one envelope per prompt subsetOpen full size ↗

AI texts · one envelope per generator configOpen full size ↗

08Top-p Outliers (Llama-3.2-3B)N3,138AUROC0.5708TPR @ 1% FPR0.0421TPR @ 0.1% FPR0.0006⌄

1% FPR threshold 0.02090.1% FPR threshold 0.0064

Threshold sweep: Top-p Outliers (Llama-3.2-3B)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Top-p Outliers (Llama-3.2-3B): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Top-p Outliers (Llama-3.2-3B): overall human

Overall · human onlyOpen full size ↗

Top-p Outliers (Llama-3.2-3B): overall AI

Overall · AI onlyOpen full size ↗

Top-p Outliers (Llama-3.2-3B): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Top-p Outliers (Llama-3.2-3B): generator subsets

AI texts · one envelope per generator configOpen full size ↗

09Entropy (Llama-3.2-3B-Instruct)N3,138AUROC0.5307TPR @ 1% FPR0.0191TPR @ 0.1% FPR0.0006⌄

1% FPR threshold 1.46480.1% FPR threshold 1.0017

Threshold sweep: Entropy (Llama-3.2-3B-Instruct)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Entropy (Llama-3.2-3B-Instruct): overall human

Overall · human onlyOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): overall AI

Overall · AI onlyOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): generator subsets

AI texts · one envelope per generator configOpen full size ↗

10Perplexity (Llama-3.2-3B-Instruct)N3,138AUROC0.5179TPR @ 1% FPR0.0153TPR @ 0.1% FPR0.0000⌄

1% FPR threshold 4.83630.1% FPR threshold 2.9954

Threshold sweep: Perplexity (Llama-3.2-3B-Instruct)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Perplexity (Llama-3.2-3B-Instruct): overall human

Overall · human onlyOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): overall AI

Overall · AI onlyOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): generator subsets

AI texts · one envelope per generator configOpen full size ↗

11Top-k Outliers (Llama-3.2-3B-Instruct)N3,138AUROC0.5044TPR @ 1% FPR0.0261TPR @ 0.1% FPR0.0000⌄

1% FPR threshold 0.03100.1% FPR threshold 0.0064

Threshold sweep: Top-k Outliers (Llama-3.2-3B-Instruct)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Top-k Outliers (Llama-3.2-3B-Instruct): overall human

Overall · human onlyOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): overall AI

Overall · AI onlyOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): generator subsets

AI texts · one envelope per generator configOpen full size ↗

12Top-p Outliers (Llama-3.2-3B-Instruct)N3,138AUROC0.5013TPR @ 1% FPR0.0389TPR @ 0.1% FPR0.0019⌄

1% FPR threshold 0.03380.1% FPR threshold 0.0232

Threshold sweep: Top-p Outliers (Llama-3.2-3B-Instruct)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Top-p Outliers (Llama-3.2-3B-Instruct): overall human

Overall · human onlyOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): overall AI

Overall · AI onlyOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): generator subsets

AI texts · one envelope per generator configOpen full size ↗

13FastDetectGPT (Llama-3.2-3B-Instruct)N3,138AUROC0.4863TPR @ 1% FPR0.0178TPR @ 0.1% FPR0.0013⌄

1% FPR threshold 1.86650.1% FPR threshold 4.4645

Threshold sweep: FastDetectGPT (Llama-3.2-3B-Instruct)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

FastDetectGPT (Llama-3.2-3B-Instruct): overall human

Overall · human onlyOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): overall AI

Overall · AI onlyOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): generator subsets

AI texts · one envelope per generator configOpen full size ↗

14Perplexity (Llama-3.2-3B)N3,138AUROC0.4837TPR @ 1% FPR0.0127TPR @ 0.1% FPR0.0006⌄

1% FPR threshold 4.02100.1% FPR threshold 1.3097

Threshold sweep: Perplexity (Llama-3.2-3B)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Perplexity (Llama-3.2-3B): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Perplexity (Llama-3.2-3B): overall human

Overall · human onlyOpen full size ↗

Overall · AI onlyOpen full size ↗

Perplexity (Llama-3.2-3B): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Perplexity (Llama-3.2-3B): generator subsets

AI texts · one envelope per generator configOpen full size ↗

15Top-k Outliers (Llama-3.2-3B)N3,138AUROC0.4734TPR @ 1% FPR0.0217TPR @ 0.1% FPR0.0006⌄

1% FPR threshold 0.02230.1% FPR threshold 0.0042

Threshold sweep: Top-k Outliers (Llama-3.2-3B)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Top-k Outliers (Llama-3.2-3B): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Top-k Outliers (Llama-3.2-3B): overall human

Overall · human onlyOpen full size ↗

Top-k Outliers (Llama-3.2-3B): overall AI

Overall · AI onlyOpen full size ↗

Top-k Outliers (Llama-3.2-3B): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

Top-k Outliers (Llama-3.2-3B): generator subsets

AI texts · one envelope per generator configOpen full size ↗

16Entropy (Llama-3.2-3B)N3,138AUROC0.4730TPR @ 1% FPR0.0096TPR @ 0.1% FPR0.0006⌄

1% FPR threshold 1.42280.1% FPR threshold 0.2395

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

Entropy (Llama-3.2-3B): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

Overall · human onlyOpen full size ↗

Overall · AI onlyOpen full size ↗

AI texts · one envelope per prompt subsetOpen full size ↗

Entropy (Llama-3.2-3B): generator subsets

AI texts · one envelope per generator configOpen full size ↗

17FastDetectGPT (Llama-3.2-3B)N3,138AUROC0.4607TPR @ 1% FPR0.0242TPR @ 0.1% FPR0.0064⌄

1% FPR threshold -2.58400.1% FPR threshold -3.4589

Threshold sweep: FastDetectGPT (Llama-3.2-3B)

TPR and FPR at every threshold · both FPR operating points markedOpen full size ↗

FastDetectGPT (Llama-3.2-3B): TPR against minimum cosdist

AI rows with cosdist below the cutoff are dropped · human rows and thresholds are unchangedOpen full size ↗

Score distributions

FastDetectGPT (Llama-3.2-3B): overall human

Overall · human onlyOpen full size ↗

FastDetectGPT (Llama-3.2-3B): overall AI

Overall · AI onlyOpen full size ↗

FastDetectGPT (Llama-3.2-3B): prompt subsets

AI texts · one envelope per prompt subsetOpen full size ↗

FastDetectGPT (Llama-3.2-3B): generator subsets

AI texts · one envelope per generator configOpen full size ↗

Model-Specific Analytics

Select a model, then a split; the split's charts appear below it. Topic × format compares human scores, AI scores and their unsigned gap; generator × prompt category shows only the AI side, since neither changes the human text; specific prompt ranks every first message by difficulty.

Select a model

01Giga EditLens Llama-3.2-3B ScoreAUROC 0.8996 · TPR@1% 0.5895⌄

Select a split · Giga EditLens Llama-3.2-3B Score

Topic × formathuman · AI · |AI − human| gridsshow ▾

Giga EditLens Llama-3.2-3B Score: Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Giga EditLens Llama-3.2-3B Score: AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Giga EditLens Llama-3.2-3B Score: Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Giga EditLens Llama-3.2-3B Score: AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

One row per distinct first message (a second turn, where there is one, is ignored), hardest first: ranked by TPR at the detector's corpus-wide 1% FPR threshold. Score columns are the detector's mean over the prompt's rows, |AI − human| is the mean per-pair score movement, and distance columns are the mean distance between each source and its rewrite. Only the 12 hardest of 287 are shown; the full ranking ↗ is a CSV beside this card.

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0067	0.0037	0.0030	0.6136	0.0600
Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0018	0.0346	0.0328	0.3317	0.0049
Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0093	0.0135	0.0042	0.0446	-0.0028
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0026	0.0671	0.0645	0.4762	0.0148
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>>. revise · single-turn	2	0.7500	0.0040	0.0064	0.0024	0.4865	0.0369
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0018	0.0036	0.0018	0.2623	0.0252
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. rewrite · single & two-turn	7	0.6939	0.0018	0.0024	0.0006	0.1034	0.0046
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>> and topic <<topic>>. rewrite · single-turn	1	1.0000	0.0036	0.0043	0.0006	0.0918	0.0217
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>>. rewrite · single-turn	2	0.7500	0.0040	0.0046	0.0006	0.0708	-0.0005
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the topic <<topic>>. rewrite · single-turn	1	0.0000	0.0027	0.0026	0.0001	0.0405	0.0042
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the format <<format>>. rewrite · single-turn	2	1.0000	0.0015	0.0304	0.0290	0.4103	0.0482
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the topic <<topic>>. rewrite · single-turn	1	1.0000	0.0030	0.0265	0.0235	0.2143	0.0876

02EditLens Llama-3.2-3B ScoreAUROC 0.8802 · TPR@1% 0.5825⌄

Select a split · EditLens Llama-3.2-3B Score

Topic × formathuman · AI · |AI − human| gridsshow ▾

EditLens Llama-3.2-3B Score: Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

EditLens Llama-3.2-3B Score: AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

EditLens Llama-3.2-3B Score: Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

EditLens Llama-3.2-3B Score: AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.2223	0.1308	0.0915	0.6136	0.0600
Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0065	0.1208	0.1143	0.3317	0.0049
Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0473	0.0862	0.0390	0.0446	-0.0028
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0083	0.1208	0.1125	0.4762	0.0148
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0190	0.0693	0.0502	0.2623	0.0252
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. rewrite · single & two-turn	7	0.5510	0.0226	0.0315	0.0091	0.1034	0.0046
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>> and topic <<topic>>. rewrite · single-turn	1	1.0000	0.0030	0.0059	0.0029	0.0918	0.0217
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>>. rewrite · single-turn	2	0.7500	0.0126	0.0166	0.0040	0.0708	-0.0005
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the topic <<topic>>. rewrite · single-turn	1	0.0000	0.0080	0.0080	0.0000	0.0405	0.0042
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the format <<format>>. rewrite · single-turn	2	1.0000	0.0066	0.0809	0.0743	0.4103	0.0482
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the topic <<topic>>. rewrite · single-turn	1	1.0000	0.0108	0.0357	0.0249	0.2143	0.0876
Edit this text by inserting a single, highly engaging introductory sentence at the very beginning, without modifying any of the original text that follows. The final text must have the topic <<topic>>. rewrite · single-turn	2	0.7500	0.0235	0.0529	0.0294	0.0541	0.0496

03Giga EditLens Roberta ScoreAUROC 0.8667 · TPR@1% 0.5545⌄

Select a split · Giga EditLens Roberta Score

Topic × formathuman · AI · |AI − human| gridsshow ▾

Giga EditLens Roberta Score: Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Giga EditLens Roberta Score: AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Giga EditLens Roberta Score: Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Giga EditLens Roberta Score: AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0071	0.0035	0.0036	0.6136	0.0600
Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0017	0.0211	0.0193	0.3317	0.0049
Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0035	0.0056	0.0021	0.0446	-0.0028
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0030	0.0758	0.0728	0.4762	0.0148
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>>. revise · single-turn	2	0.7500	0.0023	0.0040	0.0017	0.4865	0.0369
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0024	0.0028	0.0005	0.2623	0.0252
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. rewrite · single & two-turn	7	0.4490	0.0027	0.0027	0.0003	0.1034	0.0046
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>> and topic <<topic>>. rewrite · single-turn	1	0.0000	0.0025	0.0021	0.0004	0.0918	0.0217
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>>. rewrite · single-turn	2	0.2500	0.0090	0.0079	0.0011	0.0708	-0.0005
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the topic <<topic>>. rewrite · single-turn	1	0.0000	0.0019	0.0019	0.0000	0.0405	0.0042
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. rewrite · single & two-turn	7	1.0000	0.0037	0.0300	0.0263	0.3832	0.0998
Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the format <<format>>. rewrite · single-turn	2	1.0000	0.0030	0.0187	0.0157	0.4103	0.0482

04EditLens Roberta-Large ScoreAUROC 0.8550 · TPR@1% 0.4270⌄

Select a split · EditLens Roberta-Large Score

Topic × formathuman · AI · |AI − human| gridsshow ▾

EditLens Roberta-Large Score: Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

EditLens Roberta-Large Score: AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

EditLens Roberta-Large Score: Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

EditLens Roberta-Large Score: AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0269	0.2336	0.2067	0.6062	0.0759
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0124	0.1585	0.1461	0.6969	0.0590
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0475	0.0449	0.0025	0.6136	0.0600
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0612	0.2720	0.2108	0.1556	0.0673
Change the tone of this text to be more direct and bold. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.1469	0.3036	0.1567	0.6078	0.0167
Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0183	0.0592	0.0409	0.3317	0.0049
Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.1552	0.1828	0.0277	0.0446	-0.0028
Change the tone of this text to be more urgent and compelling. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0231	0.3126	0.2895	0.6758	0.0969
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0137	0.1837	0.1700	0.4762	0.0148
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>>. revise · single-turn	2	0.7500	0.0434	0.0930	0.0496	0.4865	0.0369
Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0239	0.0319	0.0080	0.2623	0.0252
Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. rewrite · single & two-turn	7	0.6122	0.0129	0.0141	0.0012	0.1034	0.0046

05LLM judge: Gemma-4-31B-it (thinking on, 8 samples)AUROC 0.7431 · TPR@1% 0.3435⌄

Select a split · LLM judge: Gemma-4-31B-it (thinking on, 8 samples)

Topic × formathuman · AI · |AI − human| gridsshow ▾

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

LLM judge: Gemma-4-31B-it (thinking on, 8 samples): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0000	0.2500	0.2500	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.0000	0.7896	0.0831
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.5000	0.0000	0.0000	0.0000	0.5860	0.0326
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0000	0.4375	0.4375	0.8067	0.2240
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.0000	0.6136	0.0600
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the topic <<topic>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.0000	0.1556	0.0673
Adapt this text for social media sharing. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0000	0.1875	0.1875	0.7609	0.0849
Change the tone of this text to be more conversational and approachable. The final text must have the topic <<topic>>. revise · single-turn	3	1.0000	0.0000	0.8333	0.8333	0.6497	0.0363
Change the tone of this text to be more direct and bold. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.2500	0.8750	0.6250	0.6078	0.0167
Change the tone of this text to be more direct and bold. The final text must have the topic <<topic>>. revise · single-turn	4	0.6250	0.0000	0.1094	0.1094	0.7222	0.0508
Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.0000	0.3317	0.0049
Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.0000	0.0446	-0.0028

06LLM judge: Qwen3.8-27B (reasoning medium, 8 samples)AUROC 0.7179 · TPR@1% 0.2811⌄

Select a split · LLM judge: Qwen3.8-27B (reasoning medium, 8 samples)

Topic × formathuman · AI · |AI − human| gridsshow ▾

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.5000	0.0000	0.0000	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	0.1875	0.1875	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.7896	0.0831
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.5000	0.0000	0.0000	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.5000	0.5000	0.8601	0.3351
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.2500	0.2500	0.8067	0.2240
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.6969	0.0590
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.6136	0.0600
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0625	0.0625	0.1556	0.0673
Adapt this text for social media sharing. The final text must have the format <<format>>. revise · single-turn	1	0.5000	0.0000	0.0000	0.7609	0.0849
Change the tone of this text to be more conversational and approachable. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.1250	0.1250	0.8135	0.1142

07BinocularsAUROC 0.5933 · TPR@1% 0.0771⌄

Select a split · Binoculars

Topic × formathuman · AI · |AI − human| gridsshow ▾

Binoculars: Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Binoculars: AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Binoculars: Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Binoculars: AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.7500	0.8675	0.8912	0.0237	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	0.8631	0.8991	0.0361	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.8320	0.8450	0.0131	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.8588	0.9104	0.0515	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.8546	0.8523	0.0344	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.2500	0.8739	0.8644	0.0140	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.8142	0.9415	0.1273	0.8601	0.3351
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.8749	0.9384	0.0635	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	0.7929	0.8241	0.0312	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.8393	0.8499	0.0106	0.6969	0.0590
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.7939	0.8031	0.0092	0.6136	0.0600
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0379	0.0645	0.0267	0.1556	0.0673

08Top-p Outliers (Llama-3.2-3B)AUROC 0.5708 · TPR@1% 0.0421⌄

Select a split · Top-p Outliers (Llama-3.2-3B)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Top-p Outliers (Llama-3.2-3B): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Top-p Outliers (Llama-3.2-3B): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Top-p Outliers (Llama-3.2-3B): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Top-p Outliers (Llama-3.2-3B): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.7500	0.0455	0.0352	0.0103	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	0.0433	0.0316	0.0117	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0261	0.0285	0.0024	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0423	0.0312	0.0110	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.0397	0.0404	0.0065	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	0.0314	0.0505	0.0191	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.5918	0.0419	0.0389	0.0095	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0520	0.0404	0.0116	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.5000	0.0388	0.0338	0.0060	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.8519	0.0442	0.0347	0.0134	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0336	0.0405	0.0069	0.6136	0.0600
Adapt this text for social media sharing. revise · single & two-turn	7	0.2245	0.0392	0.0495	0.0156	0.7721	0.1036

09Entropy (Llama-3.2-3B-Instruct)AUROC 0.5307 · TPR@1% 0.0191⌄

Select a split · Entropy (Llama-3.2-3B-Instruct)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Entropy (Llama-3.2-3B-Instruct): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Entropy (Llama-3.2-3B-Instruct): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Entropy (Llama-3.2-3B-Instruct): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	2.4637	2.6429	0.1792	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	3.0784	2.0538	1.0246	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	1.9916	1.7769	0.2147	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	3.3663	3.3212	0.0451	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	2.1969	2.2728	0.3477	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	2.4534	1.9509	0.5026	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	2.4346	2.7320	0.2974	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.5714	2.6044	2.5280	0.3357	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	3.2903	2.1371	1.1532	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	3.3277	2.3946	0.9331	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.7654	2.5134	2.1281	0.3852	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	2.6590	2.3319	0.3271	0.6969	0.0590

10Perplexity (Llama-3.2-3B-Instruct)AUROC 0.5179 · TPR@1% 0.0153⌄

Select a split · Perplexity (Llama-3.2-3B-Instruct)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Perplexity (Llama-3.2-3B-Instruct): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Perplexity (Llama-3.2-3B-Instruct): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Perplexity (Llama-3.2-3B-Instruct): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	15.0410	17.7440	2.7031	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	25.2114	10.4639	14.7475	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	10.9001	10.2232	0.6769	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	32.8996	40.9038	8.0042	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	10.6090	12.2352	2.7650	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	16.2783	9.3152	6.9631	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	11.2282	18.6152	7.3869	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.5306	16.9996	17.2901	4.8455	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	34.7585	11.6134	23.1450	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	29.4829	15.1750	14.3079	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.7901	14.9230	10.3740	4.5490	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	16.2354	12.0255	4.2098	0.6969	0.0590

11Top-k Outliers (Llama-3.2-3B-Instruct)AUROC 0.5044 · TPR@1% 0.0261⌄

Select a split · Top-k Outliers (Llama-3.2-3B-Instruct)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Top-k Outliers (Llama-3.2-3B-Instruct): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Top-k Outliers (Llama-3.2-3B-Instruct): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Top-k Outliers (Llama-3.2-3B-Instruct): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	0.1044	0.1272	0.0228	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	0.1199	0.0556	0.0643	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0684	0.0528	0.0156	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.1911	0.1696	0.0215	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.0570	0.0720	0.0248	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	0.1166	0.0625	0.0541	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	0.0773	0.1044	0.0271	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.5510	0.1139	0.1108	0.0214	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.1837	0.0505	0.1332	0.8067	0.2240
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.7778	0.0889	0.0640	0.0254	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.1035	0.0734	0.0301	0.6969	0.0590
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.1197	0.1477	0.0280	0.6136	0.0600

12Top-p Outliers (Llama-3.2-3B-Instruct)AUROC 0.5013 · TPR@1% 0.0389⌄

Select a split · Top-p Outliers (Llama-3.2-3B-Instruct)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Top-p Outliers (Llama-3.2-3B-Instruct): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Top-p Outliers (Llama-3.2-3B-Instruct): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Top-p Outliers (Llama-3.2-3B-Instruct): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.5000	0.0667	0.0547	0.0206	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.2500	0.0531	0.0617	0.0086	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0782	0.0935	0.0153	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.0684	0.0402	0.0282	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.0539	0.0526	0.0077	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.5000	0.0635	0.0651	0.0031	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	0.0386	0.0505	0.0119	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.3265	0.0520	0.0609	0.0112	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.0585	0.0707	0.0122	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.2500	0.0385	0.0418	0.0033	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.3827	0.0543	0.0598	0.0089	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0441	0.0414	0.0026	0.6969	0.0590

13FastDetectGPT (Llama-3.2-3B-Instruct)AUROC 0.4863 · TPR@1% 0.0178⌄

Select a split · FastDetectGPT (Llama-3.2-3B-Instruct)

Topic × formathuman · AI · |AI − human| gridsshow ▾

FastDetectGPT (Llama-3.2-3B-Instruct): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

FastDetectGPT (Llama-3.2-3B-Instruct): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

FastDetectGPT (Llama-3.2-3B-Instruct): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.5000	-1.9139	-1.8406	0.4436	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.2500	-1.2212	-2.9859	1.7647	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	0.0000	-3.8243	-5.0476	1.2234	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	-1.1272	-2.3619	1.2347	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	-2.1620	-2.7733	2.0793	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.5000	-2.8796	-3.2063	0.3272	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	0.1478	-1.4815	1.6293	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.3469	-1.5322	-2.2159	1.1106	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	0.0000	-2.6624	-3.3255	0.6631	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.2500	2.9460	-0.7548	3.7008	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.4074	-1.6357	-2.3984	0.9899	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	0.0000	-1.2219	-1.7350	0.5131	0.6969	0.0590

14Perplexity (Llama-3.2-3B)AUROC 0.4837 · TPR@1% 0.0127⌄

Select a split · Perplexity (Llama-3.2-3B)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Perplexity (Llama-3.2-3B): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Perplexity (Llama-3.2-3B): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Perplexity (Llama-3.2-3B): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Perplexity (Llama-3.2-3B): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	11.6376	14.8647	3.2270	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	19.3746	8.6714	10.7032	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	7.5525	6.6507	0.9017	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	25.1018	30.9267	5.8248	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	8.3577	9.5518	2.2200	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	11.9888	7.2750	4.7137	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	9.2921	16.2307	6.9387	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.4898	12.5302	14.0541	3.2892	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	24.6475	10.6846	13.9629	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	21.0244	11.1463	9.8781	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.7407	11.4873	8.6515	2.8358	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	12.6317	9.5057	3.1260	0.6969	0.0590

15Top-k Outliers (Llama-3.2-3B)AUROC 0.4734 · TPR@1% 0.0217⌄

Select a split · Top-k Outliers (Llama-3.2-3B)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Top-k Outliers (Llama-3.2-3B): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Top-k Outliers (Llama-3.2-3B): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Top-k Outliers (Llama-3.2-3B): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Top-k Outliers (Llama-3.2-3B): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	0.0822	0.0966	0.0144	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	0.0997	0.0533	0.0464	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.0391	0.0366	0.0025	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.1630	0.1518	0.0112	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.0447	0.0549	0.0185	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	0.0918	0.0512	0.0406	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	0.0700	0.1044	0.0343	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.4694	0.0924	0.1007	0.0279	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	0.1398	0.0556	0.0843	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	0.1201	0.0773	0.0429	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	0.0903	0.0527	0.0376	0.6969	0.0590
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>. revise · single-turn	1	0.0000	0.1009	0.1307	0.0298	0.6136	0.0600

16Entropy (Llama-3.2-3B)AUROC 0.4730 · TPR@1% 0.0096⌄

Select a split · Entropy (Llama-3.2-3B)

Topic × formathuman · AI · |AI − human| gridsshow ▾

Entropy (Llama-3.2-3B): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

Entropy (Llama-3.2-3B): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

Entropy (Llama-3.2-3B): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

Entropy (Llama-3.2-3B): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.2500	2.4005	2.6086	0.2081	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	1.0000	2.8746	2.2501	0.6244	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	1.0000	2.3016	2.2628	0.0388	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	3.2432	3.4026	0.1595	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	2.1256	2.1663	0.1417	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.7500	2.2854	1.9998	0.2856	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	0.0000	2.3582	2.6473	0.2891	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.4898	2.4726	2.5410	0.1999	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	1.0000	3.1396	2.4677	0.6720	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.7500	2.8847	2.3349	0.5498	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.6790	2.3850	2.1907	0.1943	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	1.0000	2.6791	2.4138	0.2653	0.6969	0.0590

17FastDetectGPT (Llama-3.2-3B)AUROC 0.4607 · TPR@1% 0.0242⌄

Select a split · FastDetectGPT (Llama-3.2-3B)

Topic × formathuman · AI · |AI − human| gridsshow ▾

FastDetectGPT (Llama-3.2-3B): Human texts · mean score per topic and format

Human texts · mean score per topic and formatOpen full size ↗

FastDetectGPT (Llama-3.2-3B): AI texts · mean score, same colour scale as the human grid

AI texts · mean score, same colour scale as the human gridOpen full size ↗

FastDetectGPT (Llama-3.2-3B): Mean unsigned |AI − human| score per pair

Mean unsigned |AI − human| score per pairOpen full size ↗

Generator × prompt categoryAI · |AI − human| gridsshow ▾

FastDetectGPT (Llama-3.2-3B): AI texts · mean score per generator config and prompt category

AI texts · mean score per generator config and prompt categoryOpen full size ↗

Mean unsigned |AI − human| score per pairOpen full size ↗

Specific prompt287 prompts · hardest firstshow ▾

Prompt (first message)	Rows	AUROC	Human score	AI score	\|AI − human\|	jaccard_1	cosdist
Adapt this text for a 5th-grade reading level. revise · two-turn	2	0.5000	0.0524	-0.2033	0.4226	0.7170	0.1166
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.0000	-0.7778	1.0650	1.8428	0.7658	0.2068
Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>. revise · single-turn	1	0.0000	2.5286	2.9996	0.4710	0.6062	0.0759
Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	0.1884	-0.1794	0.3678	0.7896	0.0831
Adapt this text for a lay audience with no technical background. revise · single & two-turn	2	0.5000	0.0610	-0.5392	1.3848	0.5927	0.0583
Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	2	0.0000	-1.4304	0.7864	2.2168	0.5860	0.0326
Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>. revise · single-turn	1	1.0000	1.2432	-1.0952	2.3384	0.8601	0.3351
Adapt this text for a teenage audience. revise · single & two-turn	7	0.4898	0.2377	0.2297	0.6339	0.7619	0.1086
Adapt this text for a teenage audience. The final text must have the format <<format>>. revise · single-turn	1	0.0000	-0.6969	0.9835	1.6805	0.8067	0.2240
Adapt this text for a teenage audience. The final text must have the topic <<topic>>. revise · single-turn	2	0.2500	0.2516	1.7789	1.5272	0.7232	0.1845
Adapt this text for an elderly adult with limited knowledge of modern day affairs. revise · single & two-turn	9	0.1235	0.0041	1.1272	1.3126	0.6889	0.1352
Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>. revise · single-turn	1	0.0000	1.3928	1.8451	0.4523	0.6969	0.0590

How far do the rewrites move?

Every distance measure between each source and its rewrite: overall, then with every prompt subset and every generator config overlaid on one chart each.

jaccard_1overall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

jaccard_2overall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

levenshteinoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

softngramoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

cosdistoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

bertscoreoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

bertscore_precisionoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

bertscore_recalloverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

moverscoreoverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

rerankeroverall · prompt subsets · generator configs⌄

OverallOpen full size ↗

One envelope per prompt subsetOpen full size ↗

One envelope per generator configOpen full size ↗

Appendix

Reference material: how the dataset was built and evaluated, its schema and loading code, univariate statistics, and cross-statistic correlations.

How the dataset was built10 generator configs · 4 prompt subsets · evaluation protocol⌄

Every row pairs original (Human) and final_response (AI): a human-written source and the text a generator model produced from it. Inclusion filtering happens while the dataset is built, so this report evaluates the published corpus exactly as it is.

Generator configs

Generator model	Temp	Sampling	Rows	Share
claude-opus-4-5-20251101	—	type disabled	162	10.3%
claude-opus-5	—	type disabled	159	10.1%
gemini-3.1-pro-preview	—	thinking_level low	162	10.3%
gemini-3.8-flash	—	thinking_level low	161	10.3%
gpt-5.4	—	reasoning_effort none	159	10.1%
gpt-5.6-sol	—	reasoning_effort none	160	10.2%
x-ai/grok-4.3	—	enabled False · max_completion_tokens 16000	145	9.2%
moonshotai/kimi-k3	—	enabled False · max_completion_tokens 16000	151	9.6%
minimax/minimax-m3	—	enabled False · max_completion_tokens 16000	151	9.6%
qwen/qwen3.7-max	—	enabled False · max_completion_tokens 16000	159	10.1%
Total			1,569	100%

Prompt subsets

Instructions are grouped by their first message. Counts cover all 1,569 rows.

direct_reference389 rows

single-turn · 60 instructions

Show the 60 instructions ▾

12× Write another piece of text that is of a similar theme or background as this text. The final text must have the format <<format>>.
11× Write a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text. The final text must have the topic <<topic>>.
11× Write another piece of text that resembles the style and language of this text. The final text must have the format <<format>> and topic <<topic>>.
10× Write a prequel to this text, describing the events, thoughts, or historical context that immediately led up to this piece being written, making sure to use similar language and wording as the author, as if it was written by the author. The final text must have the topic <<topic>>.
9× Take a subpoint within this text and write a full new text about it, using the language and wording from the text as appropriate. The final text must have the format <<format>>.
9× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should use varied sentence lengths, and remove all common AI buzzwords such as 'tapestry', 'delve', or 'testament', to ensure that it sounds like it was written by a human. The final text must have the format <<format>>.
9× Write a prequel to this text, describing the events, thoughts, or historical context that immediately led up to this piece being written, making sure to use similar language and wording as the author, as if it was written by the author.
9× Write an alternate version of this text, it should follow the style and format of the original text. The final text must have the format <<format>> and topic <<topic>>.
9× Write another piece of text that resembles the style and language of this text.
8× Take a subpoint within this text and write a full new text about it, using the language and wording from the text as appropriate.
8× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should write the new text as if it was a rough draft being typed out on a phone. The final text must have the format <<format>> and topic <<topic>>.
8× Write a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text. The final text must have the format <<format>> and topic <<topic>>.
8× Write a followup to this piece of text, in the same style and language as the text. The final text must have the format <<format>> and topic <<topic>>.
8× Write a new piece of text that is on a completely different topic, but reuses verbatim as many phrases and sentences from this text as possible. The final text must have the format <<format>> and topic <<topic>>.
8× Write a piece of text in the same style and language as this text, but is directly contrasting to the themes, ideas, and/or details in this text. The final text must have the topic <<topic>>.
8× Write another piece of text that resembles the style and language of this text. The final text must have the topic <<topic>>.
7× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should make sure there are slight, subtle grammatical mistakes, to make it sound like it was written by a human.
7× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should write the new text as if it was a rough draft being typed out on a phone.
7× Write a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text.
7× Write a detailed rebuttal or critique challenging the core ideas, themes, or facts presented in this text, using the same language and roughly matching the intellectual level of the text. The final text must have the format <<format>>.
7× Write a different piece of text on the same topic.
7× Write a different piece of text on the same topic. The final text must have the format <<format>> and topic <<topic>>.
7× Write a different piece of text on the same topic. The final text must have the format <<format>>.
7× Write a followup to this piece of text, in the same style and language as the text.
7× Write a new piece of text that is on a completely different topic, but reuses verbatim as many phrases and sentences from this text as possible. The final text must have the format <<format>>.
7× Write a new piece of text that uses this text as a citation or reference. The final text must have the format <<format>> and topic <<topic>>.
7× Write a piece of text in the same style and language as this text, but is directly contrasting to the themes, ideas, and/or details in this text. The final text must have the format <<format>> and topic <<topic>>.
7× Write a piece of text with a different topic that the author of this text might write about.
7× Write an alternate version of this text, it should follow the style and format of the original text. The final text must have the format <<format>>.
7× Write another piece of text that is of a similar theme or background as this text. The final text must have the format <<format>> and topic <<topic>>.
7× Write another piece of text that resembles the style and language of this text. The final text must have the format <<format>>.
6× Take a subpoint within this text and write a full new text about it, using the language and wording from the text as appropriate. The final text must have the topic <<topic>>.
6× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should make sure there are slight, subtle grammatical mistakes, to make it sound like it was written by a human. The final text must have the format <<format>>.
6× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should use varied sentence lengths, and remove all common AI buzzwords such as 'tapestry', 'delve', or 'testament', to ensure that it sounds like it was written by a human. The final text must have the topic <<topic>>.
6× Write a followup to this piece of text, in the same style and language as the text. The final text must have the topic <<topic>>.
6× Write a new piece of text that uses this text as a citation or reference.
6× Write a new piece of text that uses this text as a citation or reference. The final text must have the format <<format>>.
6× Write a piece of text with a different topic that the author of this text might write about. The final text must have the format <<format>>.
6× Write a prequel to this text, describing the events, thoughts, or historical context that immediately led up to this piece being written, making sure to use similar language and wording as the author, as if it was written by the author. The final text must have the format <<format>>.
6× Write an alternate version of this text, it should follow the style and format of the original text.
5× Take a subpoint within this text and write a full new text about it, using the language and wording from the text as appropriate. The final text must have the format <<format>> and topic <<topic>>.
5× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should make sure there are slight, subtle grammatical mistakes, to make it sound like it was written by a human. The final text must have the format <<format>> and topic <<topic>>.
5× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should use varied sentence lengths, and remove all common AI buzzwords such as 'tapestry', 'delve', or 'testament', to ensure that it sounds like it was written by a human. The final text must have the format <<format>> and topic <<topic>>.
5× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should write the new text as if it was a rough draft being typed out on a phone. The final text must have the format <<format>>.
5× Write a different piece of text on the same topic. The final text must have the topic <<topic>>.
5× Write a new piece of text that is on a completely different topic, but reuses verbatim as many phrases and sentences from this text as possible.
5× Write a new piece of text that is on a completely different topic, but reuses verbatim as many phrases and sentences from this text as possible. The final text must have the topic <<topic>>.
5× Write a new piece of text that uses this text as a citation or reference. The final text must have the topic <<topic>>.
5× Write a piece of text in the same style and language as this text, but is directly contrasting to the themes, ideas, and/or details in this text.
5× Write a piece of text in the same style and language as this text, but is directly contrasting to the themes, ideas, and/or details in this text. The final text must have the format <<format>>.
5× Write an alternate version of this text, it should follow the style and format of the original text. The final text must have the topic <<topic>>.
4× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should make sure there are slight, subtle grammatical mistakes, to make it sound like it was written by a human. The final text must have the topic <<topic>>.
4× Write a followup to this piece of text, in the same style and language as the text. The final text must have the format <<format>>.
4× Write a piece of text with a different topic that the author of this text might write about. The final text must have the topic <<topic>>.
4× Write another piece of text that is of a similar theme or background as this text.
3× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should use varied sentence lengths, and remove all common AI buzzwords such as 'tapestry', 'delve', or 'testament', to ensure that it sounds like it was written by a human.
3× Take inspiration from the style, language, and content of this text, and write a new piece of text on a topic of your choice. However, you should write the new text as if it was a rough draft being typed out on a phone. The final text must have the topic <<topic>>.
3× Write a prequel to this text, describing the events, thoughts, or historical context that immediately led up to this piece being written, making sure to use similar language and wording as the author, as if it was written by the author. The final text must have the format <<format>> and topic <<topic>>.
3× Write another piece of text that is of a similar theme or background as this text. The final text must have the topic <<topic>>.
2× Write a piece of text with a different topic that the author of this text might write about. The final text must have the format <<format>> and topic <<topic>>.

indirect_reference425 rows

two-turn · 25 instructions

Show the 25 instructions ▾

29× Translate the given text to German. Do not output anything besides what you were requested to write, and do not output any extra commentary.
27× Translate the given text to French. Do not output anything besides what you were requested to write, and do not output any extra commentary.
26× Translate the given text to Hindi. Do not output anything besides what you were requested to write, and do not output any extra commentary.
21× Formulate a set of strict, absolute constraints regarding format, word count, vocabulary restrictions, and syntax rules that, if followed perfectly by an AI, would result in generating a text nearly identical to this one. Note that your response should be formatted as a prompt that would be sent to an AI requesting it to generate something, not just a list of rules. Do not output anything besides what you were requested to write, and do not output any extra commentary.
21× Replace every verb in this text with the tag [VERB], keeping all nouns, adjectives, articles and punctuation exactly as they appear. Reproduce the document in full, in its original order and paragraph structure. Do not output a legend or a list of the tags you used. Do not output anything besides what you were requested to write, and do not output any extra commentary.
19× Provide only the first and last word of every sentence in this text, replacing the middle of each sentence with an ellipsis '...'. Maintain the original paragraph structure. Do not output anything besides what you were requested to write, and do not output any extra commentary.
18× Convert this text into a series of logical propositions or syllogisms that represent the core argument or narrative progression, stripped of all rhetorical flair. Do not output anything besides what you were requested to write, and do not output any extra commentary.
18× Take this text, but extract only the most meaningful sentences out of it to create a new text that is a series of disjoint sentences representing the original text. Do not output anything besides what you were requested to write, and do not output any extra commentary.
18× Translate the given text to Spanish. Do not output anything besides what you were requested to write, and do not output any extra commentary.
18× Write a detailed descriptor for the style of writing of this text, including its qualitative and quantitative properties. Do not output anything besides what you were requested to write, and do not output any extra commentary.
16× Translate the given text to Chinese. Do not output anything besides what you were requested to write, and do not output any extra commentary.
16× Write a 'few-shot' prompt using minimal, fragmented excerpts from this text as examples. The prompt should be designed to teach an LLM the pattern needed to generate the rest of the text from scratch. Do not output anything besides what you were requested to write, and do not output any extra commentary.
16× Write a detailed descriptor for how this text is stylistically differentiated from other texts, including grammatical choices and word choices not present in other texts of this kind. Do not output anything besides what you were requested to write, and do not output any extra commentary.
15× Create a prompt that might cause an LLM to generate an output resembling this text. Do not output anything besides what you were requested to write, and do not output any extra commentary.
15× Create the shortest possible prompt that can be sent to an LLM to reproduce this text. The prompt may include verbatim pieces of the original text, however, it must be short and accurate, working as an effective compression of the original text. Do not output anything besides what you were requested to write, and do not output any extra commentary.
15× Reformat this text into a sensible, structured JSON object. Do not output anything besides what you were requested to write, and do not output any extra commentary.
15× Summarize this text, attempting to preserve as much of the original language of the text as possible. Do not output anything besides what you were requested to write, and do not output any extra commentary.
15× Write a detailed descriptor for the (imaginary) personality of the author who wrote this text, including their livelihood and day to day activities, as well as how that influences their writing. Do not output anything besides what you were requested to write, and do not output any extra commentary.
14× Create a detailed dictionary that fully describes all the meaningful phrases and words this author utilizes to create this text, and how the author utilizes those words and phrases, and why. Do not output anything besides what you were requested to write, and do not output any extra commentary.
13× Take this text and replace one word in every four with a single underscore (use one underscore for each removed word, never several in a row). This way, it will be appropriate for a FIM task. You may decide which words to replace. Reproduce the rest of the document in full, in its original order and paragraph structure. Do not output anything besides what you were requested to write, and do not output any extra commentary.
13× Write a detailed descriptor for the manner in which the author of this uses language, and things such as their word choice. Do not output anything besides what you were requested to write, and do not output any extra commentary.
12× Create a prompt that would allow an LLM to generate this text as accurately/verbatim as possible, without the prompt containing any of the actual text. That is, you must use descriptors to accurately describe the text to generate. Do not output anything besides what you were requested to write, and do not output any extra commentary.
12× Envision the scenario in which the author wrote this text. Describe that scenario in exhaustive detail. Do not output anything besides what you were requested to write, and do not output any extra commentary.
12× This piece of text was cleverely generated by an LLM with a human supervising it so that the text seems authentic. Reverse engineer the process and come up with a prompt that is most probable to be the one used to generate the text. Do not output anything besides what you were requested to write, and do not output any extra commentary.
11× Translate the entirety of this text into a sequence of emojis that captures the literal meaning, tone, and narrative progression of the writing. Do not use any letters or words. Do not output anything besides what you were requested to write, and do not output any extra commentary.

revise391 rows

181 single · 210 two-turn · 119 instructions

Show the 119 instructions ▾

11× Change the tone of this text to be more inspiring and motivational.
11× Edit this text to trim the length without losing important information or tone.
10× Change the tone of this text to be more suited for a business audience.
10× Clarify this text by making the cause-and-effect relationships more obvious.
9× Adapt this text for an elderly adult with limited knowledge of modern day affairs.
9× Change the tone of this text to be more objective and unbiased.
9× Clarify this text by eliminating any ambiguous or vague language.
9× Expand this text by adding storytelling elements and dialogue to increase engagement.
9× Restructure this text to arrange the points in order of importance.
8× Change the tone of this text to be more urgent and compelling.
8× Clarify this text by making the argument and main points stronger and easier to understand.
8× Restructure this text to improve the logical flow and argument progression.
8× Restructure this text to improve the transitions and connections between ideas.
7× Adapt this text for a teenage audience.
7× Adapt this text for social media sharing.
7× Change the tone of this text to be more academic and scholarly.
7× Edit this text to eliminate redundant and filler words.
6× Change the tone of this text to be more analytical and logical.
6× Change the tone of this text to be more direct and bold.
6× Change the tone of this text to be more enthusiastic and energetic.
6× Change the tone of this text to be more persuasive and convincing.
6× Change the tone of this text to be more professional and authoritative.
6× Restructure this text to eliminate awkward phrasing and improve the overall rhythm.
6× Restructure this text to group related ideas and include clear section breaks.
5× Change the tone of this text to be more conversational and approachable.
5× Change the tone of this text to be more lighthearted and humorous.
5× Change the tone of this text to be more polite and diplomatic.
5× Clarify this text by defining any terms that might be unclear to the reader.
5× Clarify this text by paraphrasing it with different vocabulary and varied sentence structures.
5× Clarify this text by presenting the same information from a fresh angle.
5× Edit this text to be more concise and powerful.
5× Expand this text by adding backstory and context to enrich understanding.
5× Expand this text by adding concrete examples to illustrate the main points.
5× Expand this text by adding supporting information to elaborate on the key points.
5× Expand this text by adding vivid imagery and sensory details to bring it to life.
5× Restructure this text to ensure a clear introduction, body, and compelling conclusion.
4× Adapt this text for social media sharing. The final text must have the topic <<topic>>.
4× Change the tone of this text to be more direct and bold. The final text must have the topic <<topic>>.
4× Clarify this text by paraphrasing it with different vocabulary and varied sentence structures. The final text must have the format <<format>> and topic <<topic>>.
4× Expand this text by adding real-world applications and specific details.
4× Restructure this text to read more naturally, as if written by a native English speaker.
3× Change the tone of this text to be more conversational and approachable. The final text must have the topic <<topic>>.
3× Change the tone of this text to be more enthusiastic and energetic. The final text must have the format <<format>> and topic <<topic>>.
3× Change the tone of this text to be more inspiring and motivational. The final text must have the format <<format>>.
3× Change the tone of this text to be more persuasive and convincing. The final text must have the format <<format>> and topic <<topic>>.
3× Change the tone of this text to be more relaxed and friendly. The final text must have the format <<format>>.
3× Change the tone of this text to be more relaxed and friendly. The final text must have the topic <<topic>>.
3× Clarify this text by improving the phrasing while preserving its original intent.
3× Clarify this text by improving the phrasing while preserving its original intent. The final text must have the format <<format>>.
3× Restructure this text to group related ideas and include clear section breaks. The final text must have the format <<format>>.
2× Adapt this text for a 5th-grade reading level.
2× Adapt this text for a 5th-grade reading level. The final text must have the format <<format>> and topic <<topic>>.
2× Adapt this text for a lay audience with no technical background.
2× Adapt this text for a lay audience with no technical background. The final text must have the format <<format>> and topic <<topic>>.
2× Adapt this text for a teenage audience. The final text must have the topic <<topic>>.
2× Adapt this text for social media sharing. The final text must have the format <<format>> and topic <<topic>>.
2× Change the tone of this text to be more academic and scholarly. The final text must have the format <<format>> and topic <<topic>>.
2× Change the tone of this text to be more enthusiastic and energetic. The final text must have the format <<format>>.
2× Change the tone of this text to be more lighthearted and humorous. The final text must have the format <<format>> and topic <<topic>>.
2× Change the tone of this text to be more lighthearted and humorous. The final text must have the format <<format>>.
2× Change the tone of this text to be more objective and unbiased. The final text must have the format <<format>>.
2× Change the tone of this text to be more professional and authoritative. The final text must have the format <<format>> and topic <<topic>>.
2× Change the tone of this text to be more relaxed and friendly.
2× Clarify this text by making the argument and main points stronger and easier to understand. The final text must have the format <<format>> and topic <<topic>>.
2× Clarify this text by making the argument and main points stronger and easier to understand. The final text must have the topic <<topic>>.
2× Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>>.
2× Clarify this text by presenting the same information from a fresh angle. The final text must have the format <<format>>.
2× Edit this text to be more concise and powerful. The final text must have the topic <<topic>>.
2× Edit this text to eliminate redundant and filler words. The final text must have the format <<format>> and topic <<topic>>.
2× Edit this text to eliminate redundant and filler words. The final text must have the format <<format>>.
2× Expand this text by adding supporting information to elaborate on the key points. The final text must have the topic <<topic>>.
2× Restructure this text to ensure a clear introduction, body, and compelling conclusion. The final text must have the topic <<topic>>.
2× Restructure this text to group related ideas and include clear section breaks. The final text must have the topic <<topic>>.
2× Restructure this text to improve the logical flow and argument progression. The final text must have the format <<format>>.
2× Restructure this text to improve the logical flow and argument progression. The final text must have the topic <<topic>>.
2× Restructure this text to improve the transitions and connections between ideas. The final text must have the format <<format>> and topic <<topic>>.
1× Adapt this text for a 5th-grade reading level. The final text must have the format <<format>>.
1× Adapt this text for a 5th-grade reading level. The final text must have the topic <<topic>>.
1× Adapt this text for a lay audience with no technical background. The final text must have the topic <<topic>>.
1× Adapt this text for a teenage audience. The final text must have the format <<format>>.
1× Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>> and topic <<topic>>.
1× Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the format <<format>>.
1× Adapt this text for an elderly adult with limited knowledge of modern day affairs. The final text must have the topic <<topic>>.
1× Adapt this text for social media sharing. The final text must have the format <<format>>.
1× Change the tone of this text to be more academic and scholarly. The final text must have the topic <<topic>>.
1× Change the tone of this text to be more analytical and logical. The final text must have the topic <<topic>>.
1× Change the tone of this text to be more conversational and approachable. The final text must have the format <<format>> and topic <<topic>>.
1× Change the tone of this text to be more conversational and approachable. The final text must have the format <<format>>.
1× Change the tone of this text to be more direct and bold. The final text must have the format <<format>> and topic <<topic>>.
1× Change the tone of this text to be more empathetic and understanding.
1× Change the tone of this text to be more empathetic and understanding. The final text must have the format <<format>>.
1× Change the tone of this text to be more objective and unbiased. The final text must have the topic <<topic>>.
1× Change the tone of this text to be more persuasive and convincing. The final text must have the topic <<topic>>.
1× Change the tone of this text to be more polite and diplomatic. The final text must have the format <<format>>.
1× Change the tone of this text to be more professional and authoritative. The final text must have the topic <<topic>>.
1× Change the tone of this text to be more relaxed and friendly. The final text must have the format <<format>> and topic <<topic>>.
1× Change the tone of this text to be more urgent and compelling. The final text must have the format <<format>> and topic <<topic>>.
1× Change the tone of this text to be more urgent and compelling. The final text must have the format <<format>>.
1× Clarify this text by defining any terms that might be unclear to the reader. The final text must have the format <<format>>.
1× Clarify this text by eliminating any ambiguous or vague language. The final text must have the format <<format>> and topic <<topic>>.
1× Clarify this text by making the argument and main points stronger and easier to understand. The final text must have the format <<format>>.
1× Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the format <<format>> and topic <<topic>>.
1× Clarify this text by making the cause-and-effect relationships more obvious. The final text must have the topic <<topic>>.
1× Clarify this text by paraphrasing it with different vocabulary and varied sentence structures. The final text must have the topic <<topic>>.
1× Clarify this text by presenting the same information from a fresh angle. The final text must have the format <<format>> and topic <<topic>>.
1× Clarify this text by presenting the same information from a fresh angle. The final text must have the topic <<topic>>.
1× Edit this text to eliminate redundant and filler words. The final text must have the topic <<topic>>.
1× Expand this text by adding backstory and context to enrich understanding. The final text must have the format <<format>> and topic <<topic>>.
1× Expand this text by adding concrete examples to illustrate the main points. The final text must have the format <<format>> and topic <<topic>>.
1× Expand this text by adding storytelling elements and dialogue to increase engagement. The final text must have the format <<format>> and topic <<topic>>.
1× Expand this text by adding storytelling elements and dialogue to increase engagement. The final text must have the topic <<topic>>.
1× Expand this text by adding supporting information to elaborate on the key points. The final text must have the format <<format>> and topic <<topic>>.
1× Restructure this text to arrange the points in order of importance. The final text must have the format <<format>>.
1× Restructure this text to arrange the points in order of importance. The final text must have the topic <<topic>>.
1× Restructure this text to eliminate awkward phrasing and improve the overall rhythm. The final text must have the format <<format>> and topic <<topic>>.
1× Restructure this text to eliminate awkward phrasing and improve the overall rhythm. The final text must have the topic <<topic>>.
1× Restructure this text to group related ideas and include clear section breaks. The final text must have the format <<format>> and topic <<topic>>.
1× Restructure this text to improve the logical flow and argument progression. The final text must have the format <<format>> and topic <<topic>>.
1× Restructure this text to read more naturally, as if written by a native English speaker. The final text must have the format <<format>>.

rewrite364 rows

179 single · 185 two-turn · 83 instructions

Show the 83 instructions ▾

15× Edit this text by replacing all existing transition words and phrases (e.g., 'however', 'therefore', 'in addition') with different, synonymous transitions, making no other changes.
15× Modify this text so that it is as short as possible. At least a quarter of the original text must be repeated verbatim, and at most half of it; the remainder must be new prose.
14× Edit this text by replacing all instances of passive voice with active voice, without altering any sentences that are already in the active voice.
12× Edit this text by removing all adverbs, making only minimal adjustments to the surrounding grammar to ensure the sentences remain structurally sound.
12× Edit this text by splitting the three longest sentences into shorter, distinct sentences, without changing the vocabulary or structure of the rest of the text.
10× Edit this text by inserting twenty to thirty words (in appropriate locations), without changing anything else.
10× Edit this text by upgrading the verbs to be more dynamic and action-oriented, leaving the subjects, objects, and overall length completely unchanged.
9× Edit this text by inserting a single, highly engaging introductory sentence at the very beginning, without modifying any of the original text that follows.
9× Edit this text by inserting a singular new section (in an appropriate locations), you may only make minimal modifications to transitions in adjacent sections for better flow.
9× Edit this text by inserting exactly one to three sentences (in appropriate locations), you may only make minimal modifications to transitions in adjacent sentences for better flow.
9× Edit this text by removing five to ten percent of the words from it (the least necessary), without changing anything else.
9× Edit this text by replacing exactly one to three sentences, without changing anything else.
9× Edit this text by replacing five to ten percent of words in it, without changing anything else.
9× Edit this text by replacing generic adjectives with highly specific, evocative ones, modifying no more than ten percent of the total words.
9× Modify this text to be as long as possible. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose.
9× Rewrite exactly one section of this text so that its wording is entirely new, and leave every other section untouched. The rewritten section must share no sentence with the original.
9× Rewrite only the opening hook and the final concluding sentence of this text, leaving the entire middle portion exactly as it was originally written.
9× Rewrite the weakest section in this text, but do not make any modifications to any other section.
7× Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched.
7× Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim.
7× Edit this text by inserting a singular new section (in an appropriate locations), you may only make minimal modifications to transitions in adjacent sections for better flow. The final text must have the format <<format>> and topic <<topic>>.
7× Edit this text by inserting twenty to thirty words (in appropriate locations), without changing anything else. The final text must have the format <<format>> and topic <<topic>>.
7× Edit this text by removing one section from it (the least necessary), you may only make minimal modifications to adjacent sections for better flow.
6× Edit this text by replacing the final sentence with a more open-ended, thought-provoking statement, without altering any preceding text.
6× Edit this text by replacing, removing or inserting exactly one to three sentences, without changing anything else.
5× Edit this text by inserting a single, highly engaging introductory sentence at the very beginning, without modifying any of the original text that follows. The final text must have the format <<format>>.
5× Edit this text by removing exactly one to three sentences from it (the least necessary), you may only make minimal modifications to adjacent sentences for better flow.
5× Edit this text by replacing the final sentence with a more open-ended, thought-provoking statement, without altering any preceding text. The final text must have the format <<format>>.
5× Improve this text by rewriting it. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose.
5× Rewrite only the opening hook and the final concluding sentence of this text, leaving the entire middle portion exactly as it was originally written. The final text must have the topic <<topic>>.
4× Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the format <<format>> and topic <<topic>>.
4× Edit this text by inserting a single, highly engaging introductory sentence at the very beginning, without modifying any of the original text that follows. The final text must have the format <<format>> and topic <<topic>>.
4× Edit this text by replacing all instances of passive voice with active voice, without altering any sentences that are already in the active voice. The final text must have the format <<format>> and topic <<topic>>.
4× Edit this text by splitting the three longest sentences into shorter, distinct sentences, without changing the vocabulary or structure of the rest of the text. The final text must have the topic <<topic>>.
4× Rewrite the weakest section in this text, but do not make any modifications to any other section. The final text must have the format <<format>> and topic <<topic>>.
4× Rewrite the weakest section in this text, but do not make any modifications to any other section. The final text must have the format <<format>>.
3× Edit this text by inserting exactly one to three sentences (in appropriate locations), you may only make minimal modifications to transitions in adjacent sentences for better flow. The final text must have the format <<format>> and topic <<topic>>.
3× Edit this text by removing one section from it (the least necessary), you may only make minimal modifications to adjacent sections for better flow. The final text must have the format <<format>> and topic <<topic>>.
3× Edit this text by removing one section from it (the least necessary), you may only make minimal modifications to adjacent sections for better flow. The final text must have the format <<format>>.
3× Edit this text by replacing five to ten percent of words in it, without changing anything else. The final text must have the format <<format>> and topic <<topic>>.
3× Edit this text by replacing the final sentence with a more open-ended, thought-provoking statement, without altering any preceding text. The final text must have the topic <<topic>>.
3× Edit this text by replacing, removing or inserting twenty to thirty words, without changing anything else.
3× Edit this text by replacing, removing or inserting twenty to thirty words, without changing anything else. The final text must have the format <<format>>.
3× Edit this text by upgrading the verbs to be more dynamic and action-oriented, leaving the subjects, objects, and overall length completely unchanged. The final text must have the format <<format>> and topic <<topic>>.
3× Modify this text so that it is as short as possible. At least a quarter of the original text must be repeated verbatim, and at most half of it; the remainder must be new prose. The final text must have the format <<format>> and topic <<topic>>.
2× Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>>.
2× Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the format <<format>>.
2× Edit this text by inserting a single, highly engaging introductory sentence at the very beginning, without modifying any of the original text that follows. The final text must have the topic <<topic>>.
2× Edit this text by removing exactly one to three sentences from it (the least necessary), you may only make minimal modifications to adjacent sentences for better flow. The final text must have the format <<format>> and topic <<topic>>.
2× Edit this text by removing five to ten percent of the words from it (the least necessary), without changing anything else. The final text must have the format <<format>> and topic <<topic>>.
2× Edit this text by replacing five to ten percent of words in it, without changing anything else. The final text must have the topic <<topic>>.
2× Edit this text by replacing generic adjectives with highly specific, evocative ones, modifying no more than ten percent of the total words. The final text must have the format <<format>>.
2× Edit this text by splitting the three longest sentences into shorter, distinct sentences, without changing the vocabulary or structure of the rest of the text. The final text must have the format <<format>> and topic <<topic>>.
2× Edit this text by splitting the three longest sentences into shorter, distinct sentences, without changing the vocabulary or structure of the rest of the text. The final text must have the format <<format>>.
2× Improve this text by rewriting it. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose. The final text must have the format <<format>> and topic <<topic>>.
2× Improve this text by rewriting it. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose. The final text must have the format <<format>>.
2× Improve this text by rewriting it. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose. The final text must have the topic <<topic>>.
2× Modify this text so that it is as short as possible. At least a quarter of the original text must be repeated verbatim, and at most half of it; the remainder must be new prose. The final text must have the topic <<topic>>.
2× Modify this text to be as long as possible. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose. The final text must have the format <<format>> and topic <<topic>>.
2× Rewrite exactly one section of this text so that its wording is entirely new, and leave every other section untouched. The rewritten section must share no sentence with the original. The final text must have the format <<format>> and topic <<topic>>.
2× Rewrite the weakest section in this text, but do not make any modifications to any other section. The final text must have the topic <<topic>>.
1× Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the format <<format>> and topic <<topic>>.
1× Edit this text by combining at least four pairs of short, adjacent sentences into complex sentences using appropriate conjunctions, leaving all other sentences untouched. The final text must have the topic <<topic>>.
1× Edit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim. The final text must have the topic <<topic>>.
1× Edit this text by inserting a singular new section (in an appropriate locations), you may only make minimal modifications to transitions in adjacent sections for better flow. The final text must have the topic <<topic>>.
1× Edit this text by inserting twenty to thirty words (in appropriate locations), without changing anything else. The final text must have the format <<format>>.
1× Edit this text by inserting twenty to thirty words (in appropriate locations), without changing anything else. The final text must have the topic <<topic>>.
1× Edit this text by removing all adverbs, making only minimal adjustments to the surrounding grammar to ensure the sentences remain structurally sound. The final text must have the format <<format>> and topic <<topic>>.
1× Edit this text by removing five to ten percent of the words from it (the least necessary), without changing anything else. The final text must have the format <<format>>.
1× Edit this text by replacing all existing transition words and phrases (e.g., 'however', 'therefore', 'in addition') with different, synonymous transitions, making no other changes. The final text must have the format <<format>> and topic <<topic>>.
1× Edit this text by replacing all existing transition words and phrases (e.g., 'however', 'therefore', 'in addition') with different, synonymous transitions, making no other changes. The final text must have the topic <<topic>>.
1× Edit this text by replacing all instances of passive voice with active voice, without altering any sentences that are already in the active voice. The final text must have the format <<format>>.
1× Edit this text by replacing all instances of passive voice with active voice, without altering any sentences that are already in the active voice. The final text must have the topic <<topic>>.
1× Edit this text by replacing exactly one to three sentences, without changing anything else. The final text must have the format <<format>> and topic <<topic>>.
1× Edit this text by replacing exactly one to three sentences, without changing anything else. The final text must have the format <<format>>.
1× Edit this text by replacing five to ten percent of words in it, without changing anything else. The final text must have the format <<format>>.
1× Edit this text by replacing generic adjectives with highly specific, evocative ones, modifying no more than ten percent of the total words. The final text must have the topic <<topic>>.
1× Edit this text by replacing the final sentence with a more open-ended, thought-provoking statement, without altering any preceding text. The final text must have the format <<format>> and topic <<topic>>.
1× Edit this text by replacing, removing or inserting exactly one to three sentences, without changing anything else. The final text must have the topic <<topic>>.
1× Edit this text by replacing, removing or inserting twenty to thirty words, without changing anything else. The final text must have the topic <<topic>>.
1× Edit this text by upgrading the verbs to be more dynamic and action-oriented, leaving the subjects, objects, and overall length completely unchanged. The final text must have the format <<format>>.
1× Edit this text by upgrading the verbs to be more dynamic and action-oriented, leaving the subjects, objects, and overall length completely unchanged. The final text must have the topic <<topic>>.
1× Modify this text to be as long as possible. At least half of the original text must be repeated verbatim, and at most three quarters of it; the remainder must be new prose. The final text must have the topic <<topic>>.

Topics

Topic	Rows	Share
Finance & Business	120	7.6%
Sports & Fitness	100	6.4%
Hardware	86	5.5%
Religion	83	5.3%
Social Life	80	5.1%
Software	80	5.1%
Food & Dining	77	4.9%
Health	77	4.9%
Literature	77	4.9%
Education & Jobs	76	4.8%
Science & Tech.	76	4.8%
Travel	74	4.7%
Politics	73	4.7%
Software Dev.	73	4.7%
Transportation	67	4.3%
Home & Hobbies	62	4.0%
Crime & Law	57	3.6%
Entertainment	47	3.0%
History	47	3.0%
Industrial	44	2.8%
Games	43	2.7%
Art & Design	34	2.2%
Fashion & Beauty	16	1.0%

Formats

Format	Rows	Share
Personal Blog	290	18.5%
News Article	261	16.6%
Nonfiction Writing	114	7.3%
Tutorial	64	4.1%
Product Page	60	3.8%
News (Org.)	59	3.8%
Knowledge Article	58	3.7%
Listicle	55	3.5%
User Review	54	3.4%
About (Org.)	51	3.3%
Academic Writing	49	3.1%
Customer Support	48	3.1%
Q&A Forum	48	3.1%
Documentation	47	3.0%
Legal Notices	44	2.8%
FAQ	43	2.7%
Structured Data	42	2.7%
Creative Writing	41	2.6%
Comment Section	39	2.5%
About (Pers.)	36	2.3%
Audio Transcript	36	2.3%
Content Listing	30	1.9%

Evaluation protocol

Classes

original is Human, final_response is AI. Each detector scores both texts of every row.

Thresholds

All 1,569 rows are used; there is no sweep/evaluation split. Each detector's thresholds are placed exactly on its human scores so that 1% and 0.1% of human texts are called AI. 9 detectors call lower scores AI (for example, perplexity).

Metrics

N (texts scored), AUROC, and TPR at the 1% and 0.1% FPR thresholds. Subsets are scored at the corpus-wide thresholds rather than refitted.

Filters

None. No cosine, soft-ngram, Jaccard, or other analysis-time filtering is applied.

Skipped

None

Config

config/analysis_llmjudge_mini.toml

Schema & usage65 columns · loading snippet⌄

65 columns per row. Detector columns come in pairs, one per text of the row.

Text

original

Human-written source (class: Human)

final_response

Generated text (class: AI)

response_0

Generator output for the first turn

response_1

Generator output for the second turn (empty for single-turn prompts)

Generation

prompt

struct: chat_turns (list of strings), use_multiturn, examples, metadata.PROMPT_TYPE

generator_model

Hugging Face id of the generator model

generation_params

JSON string of sampling parameters

Metadata

topic

Topic of the human source

format

Format of the human source

Distances

jaccard_1, jaccard_2, levenshtein, softngram, cosdist, bertscore, bertscore_precision, bertscore_recall, moverscore, reranker

10 float columns measuring how far the rewrite moved from the original

Other columns

original_editlens_bucket_roberta_large, final_response_editlens_bucket_roberta_large, original_editlens_bucket_llama_3_2_3b, final_response_editlens_bucket_llama_3_2_3b, original_editlens_bucket_giga_roberta, final_response_editlens_bucket_giga_roberta, original_editlens_bucket_giga_llama_3_2_3b, final_response_editlens_bucket_giga_llama_3_2_3b, original_llmjudge_valid_qwen38_27b, final_response_llmjudge_valid_qwen38_27b, original_llmjudge_valid_gemma4_31b, final_response_llmjudge_valid_gemma4_31b

Not used by this report

Detector scores

One float column per text: {original|final_response}<suffix>

_editlens_score_roberta_large

EditLens Roberta-Large Score

_editlens_score_llama_3_2_3b

EditLens Llama-3.2-3B Score

_editlens_score_giga_roberta

Giga EditLens Roberta Score

_editlens_score_giga_llama_3_2_3b

Giga EditLens Llama-3.2-3B Score

_perplexity_llama_instruct

Perplexity (Llama-3.2-3B-Instruct)

_perplexity_llama_base

Perplexity (Llama-3.2-3B)

_entropy_llama_instruct

Entropy (Llama-3.2-3B-Instruct)

_entropy_llama_base

Entropy (Llama-3.2-3B)

_topp_outlier_llama_instruct

Top-p Outliers (Llama-3.2-3B-Instruct)

_topp_outlier_llama_base

Top-p Outliers (Llama-3.2-3B)

_topk_outlier_llama_instruct

Top-k Outliers (Llama-3.2-3B-Instruct)

_topk_outlier_llama_base

Top-k Outliers (Llama-3.2-3B)

_fastdetectgpt_llama_instruct

FastDetectGPT (Llama-3.2-3B-Instruct)

_fastdetectgpt_llama_base

FastDetectGPT (Llama-3.2-3B)

_binoculars

Binoculars

_llmjudge_score_qwen38_27b

LLM judge: Qwen3.8-27B (reasoning medium, 8 samples)

_llmjudge_score_gemma4_31b

LLM judge: Gemma-4-31B-it (thinking on, 8 samples)

Load it

from datasets import concatenate_datasets, get_dataset_config_names, load_dataset

repo = "G-reen/fastdetector-test-stat-test-mini"
# every generator config is its own dataset config
ds = concatenate_datasets([
    load_dataset(repo, config, split="train")
    for config in get_dataset_config_names(repo)
])

Univariate statistics44 statistics · all 1,569 rows valid⌄

Every statistic the report does arithmetic on, over the complete corpus. Missing or non-finite values are counted as invalid.

Statistic	Mean	Median	Std	Min	Max
`jaccard_1`	0.5733	0.6792	0.2841	0.0169	0.9839
`jaccard_2`	0.6920	0.8427	0.3128	0.0229	1.0000
`levenshtein`	2083.6922	1527.0000	2280.9246	3.0000	32276.0000
`softngram`	0.5244	0.5653	0.3636	0.0000	1.0000
`cosdist`	0.1542	0.0856	0.1728	-0.0062	0.8306
`bertscore`	0.1177	0.1252	0.0726	0.0018	0.3631
`bertscore_precision`	0.1160	0.1220	0.0710	0.0020	0.3683
`bertscore_recall`	0.1186	0.1213	0.0781	0.0013	0.3579
`moverscore`	0.4846	0.5402	0.1998	0.0809	0.9746
`reranker`	-6.0039	-7.3125	3.8663	-10.5625	14.6250
`EditLens Roberta-Large Score (Human)`	0.0452	0.0204	0.0713	0.0064	0.7948
`EditLens Roberta-Large Score (AI)`	0.3619	0.2534	0.3439	0.0072	0.9996
`EditLens Llama-3.2-3B Score (Human)`	0.0286	0.0155	0.0371	0.0005	0.3730
`EditLens Llama-3.2-3B Score (AI)`	0.3470	0.2843	0.3149	0.0007	1.0000
`Giga EditLens Roberta Score (Human)`	0.0089	0.0038	0.0298	0.0013	0.6720
`Giga EditLens Roberta Score (AI)`	0.3861	0.2339	0.4020	0.0013	1.0000
`Giga EditLens Llama-3.2-3B Score (Human)`	0.0052	0.0027	0.0130	0.0005	0.1835
`Giga EditLens Llama-3.2-3B Score (AI)`	0.3339	0.1830	0.3666	0.0008	1.0000
`Perplexity (Llama-3.2-3B-Instruct) (Human)`	15.1916	13.6999	6.9329	2.9954	64.9892
`Perplexity (Llama-3.2-3B-Instruct) (AI)`	15.2946	13.5247	12.1957	3.1445	344.7711
`Perplexity (Llama-3.2-3B) (Human)`	11.5582	10.5363	4.9229	1.2068	46.1120
`Perplexity (Llama-3.2-3B) (AI)`	12.3259	10.8865	9.8122	1.3097	280.8260
`Entropy (Llama-3.2-3B-Instruct) (Human)`	2.4694	2.4648	0.4319	1.0731	4.2297
`Entropy (Llama-3.2-3B-Instruct) (AI)`	2.4219	2.4102	0.4770	1.0017	4.3832
`Entropy (Llama-3.2-3B) (Human)`	2.3592	2.3616	0.3807	0.2294	3.7963
`Entropy (Llama-3.2-3B) (AI)`	2.3954	2.3920	0.4005	0.2395	3.9796
`Top-p Outliers (Llama-3.2-3B-Instruct) (Human)`	0.0550	0.0533	0.0122	0.0235	0.1231
`Top-p Outliers (Llama-3.2-3B-Instruct) (AI)`	0.0553	0.0533	0.0145	0.0206	0.1750
`Top-p Outliers (Llama-3.2-3B) (Human)`	0.0416	0.0412	0.0096	0.0064	0.0796
`Top-p Outliers (Llama-3.2-3B) (AI)`	0.0396	0.0386	0.0122	0.0064	0.1333
`Top-k Outliers (Llama-3.2-3B-Instruct) (Human)`	0.0934	0.0896	0.0344	0.0064	0.2358
`Top-k Outliers (Llama-3.2-3B-Instruct) (AI)`	0.0938	0.0894	0.0411	0.0101	0.5000
`Top-k Outliers (Llama-3.2-3B) (Human)`	0.0753	0.0710	0.0304	0.0039	0.2141
`Top-k Outliers (Llama-3.2-3B) (AI)`	0.0796	0.0745	0.0375	0.0042	0.3467
`FastDetectGPT (Llama-3.2-3B-Instruct) (Human)`	-1.8762	-1.7862	1.6299	-10.2423	5.9897
`FastDetectGPT (Llama-3.2-3B-Instruct) (AI)`	-1.9506	-1.8828	1.7872	-12.6989	5.8664
`FastDetectGPT (Llama-3.2-3B) (Human)`	-0.0811	-0.0718	1.0324	-4.6137	2.7563
`FastDetectGPT (Llama-3.2-3B) (AI)`	0.1391	0.0551	1.4862	-4.3856	11.1712
`Binoculars (Human)`	0.8473	0.8551	0.0695	0.0379	1.0063
`Binoculars (AI)`	0.8680	0.8672	0.0676	0.0645	1.2509
`LLM judge: Qwen3.8-27B (reasoning medium, 8 samples) (Human)`	0.0118	0.0000	0.0822	0.0000	0.9375
`LLM judge: Qwen3.8-27B (reasoning medium, 8 samples) (AI)`	0.2935	0.0000	0.3961	0.0000	1.0000
`LLM judge: Gemma-4-31B-it (thinking on, 8 samples) (Human)`	0.0272	0.0000	0.1393	0.0000	1.0000
`LLM judge: Gemma-4-31B-it (thinking on, 8 samples) (AI)`	0.4357	0.1250	0.4623	0.0000	1.0000

Correlation heatmappairwise Pearson r⌄

Each pair is correlated over the rows where both are finite.Open full size ↗

Generated by the FastDetector analysis pipeline from the complete published corpus.config/analysis_llmjudge_mini.toml

Downloads last month: 350