Datasets:

G-reen
/

cc-2021-stat

original stringlengths 171 44.4k	prompt dict	response_0 stringlengths 65 15.9k	response_1 stringlengths 0 15.6k	final_response stringlengths 0 15.6k	generator_model stringclasses 1 value	generation_params stringclasses 1 value	jaccard_1 float64 0 1	jaccard_2 float64 0 1	levenshtein float64 0 37.6k	softngram float64 0 1	cosdist float64 -0.01 0.83	bertscore_precision float64 0 1	bertscore_recall float64 0 1	bertscore float64 0 1	moverscore float64 0.03 1	reranker float64 -10.81 14.6	original_editlens_bucket_roberta_large int64 0 3	original_editlens_score_roberta_large float64 0.01 0.88	final_response_editlens_bucket_roberta_large int64 0 3	final_response_editlens_score_roberta_large float64 0.01 1	original_perplexity_llama_instruct float64 1.74 675	original_entropy_llama_instruct float64 0.65 7.09	original_topp_outlier_llama_instruct float64 0 0.15	original_topk_outlier_llama_instruct float64 0 0.56	original_fastdetectgpt_llama_instruct float64 -10.81 13.5	original_perplexity_llama_base float64 1.15 404	original_entropy_llama_base float64 0.23 5.88	original_topp_outlier_llama_base float64 0 0.12	original_topk_outlier_llama_base float64 0 0.53	original_fastdetectgpt_llama_base float64 -7.27 2.75	original_binoculars float64 0.05 1.18	final_response_perplexity_llama_instruct float64 1.25 331 ⌀	final_response_entropy_llama_instruct float64 0 5.84	final_response_topp_outlier_llama_instruct float64 0.01 0.17 ⌀	final_response_topk_outlier_llama_instruct float64 0.01 0.43 ⌀	final_response_fastdetectgpt_llama_instruct float64 -6.99 12.2	final_response_perplexity_llama_base float64 1.27 275 ⌀	final_response_entropy_llama_base float64 0 4.83	final_response_topp_outlier_llama_base float64 0 0.14 ⌀	final_response_topk_outlier_llama_base float64 0 0.41 ⌀	final_response_fastdetectgpt_llama_base float64 -5.27 4.13	final_response_binoculars float64 0 1.13
Thank you for sharing this sad update. It began early in June. Diego would leave the litter box with a wet back end. I'd have to try to catch him and dry him off. He was also being aggressive towards Cali so after consultation with the... Had Cali at the vet for a health check and shots. Also had the vet look at the tw...	{ "chat_turns": [ "{{DOC}}\nChange the tone of this text to be more relaxed and friendly.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "PROMPT_TYPE": "revise" } }	Hey there! 😺 It’s been a wild June with Diego leaving the litter box a little wet, but we’ve got him dried off and back to his usual self. He’s also been a bit grumpy with Cali, but after chatting with the vet, things are looking up! Cali had a great check-up and got her shots. The vet checked out those two lumps—tu...		Hey there! 😺 It’s been a wild June with Diego leaving the litter box a little wet, but we’ve got him dried off and back to his usual self. He’s also been a bit grumpy with Cali, but after chatting with the vet, things are looking up! Cali had a great check-up and got her shots. The vet checked out those two lumps—tu...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.781931	0.936082	947	0.694805	0.28504	0.167541	0.147114	0.157451	0.594106	-4.4375	0	0.009655	1	0.442052	29.431168	2.968705	0.052308	0.132308	-3.099721	21.108406	3.051931	0.046154	0.107692	0.017457	0.912657	16.330864	2.353805	0.082888	0.104278	-4.099265	13.250969	2.704551	0.042781	0.090909	1.065956	0.918024
CLXVI Sanjaya said, Karna, the son of Vikartana O king, resisted the mighty car-warrior Sahadeva in that battle, who advanced from desire of getting at Drona. Piercing the son of Radha with nine shafts, Sahadeva once more pierced that warrior with nine straight arrows. Karna then pierced Sahadeva in return with a hundr...	{ "chat_turns": [ "{{DOC}}\nModify this text to be as long as possible. However, the new text must contain at least half of the original text, unmodified, repeated verbatim.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "...	CLXVI Sanjaya said, Karna, the son of Vikartana O king, resisted the mighty car-warrior Sahadeva in that battle, who advanced from desire of getting at Drona. Piercing the son of Radha with nine shafts, Sahadeva once more pierced that warrior with nine straight arrows. Karna then pierced Sahadeva in return with a hundr...		CLXVI Sanjaya said, Karna, the son of Vikartana O king, resisted the mighty car-warrior Sahadeva in that battle, who advanced from desire of getting at Drona. Piercing the son of Radha with nine shafts, Sahadeva once more pierced that warrior with nine straight arrows. Karna then pierced Sahadeva in return with a hundr...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.004032	0.006726	1,104	0.003417	0.018178	0.001315	0.001286	0.001301	0.077886	-5	0	0.023015	0	0.023015	14.173077	2.463246	0.05227	0.115543	-2.228704	8.895827	2.265456	0.035763	0.07978	0.995518	0.754622	7.115674	1.813219	0.040959	0.083916	-2.400682	5.113295	1.67952	0.030969	0.058941	0.805168	0.762423
This processor is one of the most famous one in guitar racks and old studios. It was introduced in 1985 and quickly became successful, it was relatively cheap, and has some very interesting effects algorithms. One of them is the famous ‘Pitch Change C’ algorithm, used by many legendary session guitarists in that era, i...	{ "chat_turns": [ "{{DOC}}\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 write the new text as if it was a rough draft being typed out on a phone.\nOutput the full new text with no extra statements or commentation...	I recently acquired a vintage synthesizer from the early 1990s that had been sitting in storage for decades. It was a Roland Juno-106, known for its lush analog sound and iconic patch memory. Introduced in 1983, it quickly became a staple in both home studios and professional music production. Its affordability and ric...		I recently acquired a vintage synthesizer from the early 1990s that had been sitting in storage for decades. It was a Roland Juno-106, known for its lush analog sound and iconic patch memory. Introduced in 1983, it quickly became a staple in both home studios and professional music production. Its affordability and ric...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.86747	0.958462	2,274	0.944058	0.407494	0.218804	0.175631	0.197798	0.675692	1.1875	0	0.013802	3	0.99845	30.019443	3.32871	0.046099	0.131206	-0.487966	22.984729	3.058342	0.049645	0.113475	-0.5479	0.846486	5.703372	1.696582	0.036395	0.020797	-0.610455	5.377899	1.828387	0.025997	0.025997	1.984839	0.842412
This post explains the difference between foodservice operations that are outsourced to contract management companies and those that are 'self-operated' (self-op). You’re definitely going to want to sell to these folks. It’s not an either/or, self-op or contract-managed, situation. You’re going to want to go after both...	{ "chat_turns": [ "{{DOC}}\nChange the tone of this text to be more empathetic and understanding.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "PROMPT_TYPE": "revise" } }	Understanding the nuances between outsourced foodservice operations managed by contract management companies (CFMs) and self-operated (self-op) programs is crucial for tailoring your sales strategy effectively. It’s not an either/or choice; ideally, you’ll want to engage with both types of institutions. Recognizing the...		Understanding the nuances between outsourced foodservice operations managed by contract management companies (CFMs) and self-operated (self-op) programs is crucial for tailoring your sales strategy effectively. It’s not an either/or choice; ideally, you’ll want to engage with both types of institutions. Recognizing the...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.763006	0.941235	2,751	0.731161	0.112834	0.182736	0.149513	0.166456	0.622445	-6.375	1	0.301078	3	0.99954	12.161763	2.362016	0.057598	0.072304	-1.86494	9.853105	2.438624	0.033088	0.061275	2.1153	0.877299	9.850228	2.105904	0.047847	0.062201	-2.370973	8.815611	2.121711	0.036683	0.060606	-0.739486	0.910723
Personalized baskets filled with a selection of foods, personal care products and specialty gifts, hand-crafted in Maine. Biologics, or drugs derived from living cells, often are deemed to be specialty drugs. In the Extensions tab, SDKs that apply to the project from which the Reference Manager dialog box was invoked a...	{ "chat_turns": [ "{{DOC}}\nWrite a detailed descriptor for the style of writing of this text, including its qualitative and quantitative properties.\nDo not output anything besides what you were requested to write, and do not output any extra commentary.", "{{RESP_0}}\nAbove is an AI generated descriptor/tra...	The style of writing in this text can be characterized as follows: Qualitative Properties: 1. Mixed and Varied Content: The text combines disparate topics—personalized baskets, biologics, software development SDKs, a socio-economic scenario from Annawadi, navigation features, health-related queries (blood pres...	You're right that many details have been omitted, but I can reconstruct a plausible version of the original text based on the qualitative and quantitative properties described. Here's a recreated text that captures the essence of the descriptor: --- Hey, you're looking for some personalized baskets filled with goodie...	You're right that many details have been omitted, but I can reconstruct a plausible version of the original text based on the qualitative and quantitative properties described. Here's a recreated text that captures the essence of the descriptor: --- Hey, you're looking for some personalized baskets filled with goodie...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.902311	0.986193	5,601	0.96901	0.494667	0.194548	0.197042	0.195797	0.682482	-0.25	0	0.015037	2	0.629044	43.871325	3.602982	0.055921	0.221053	-2.661363	34.701647	3.389289	0.058553	0.198684	-2.479291	0.902964	11.367651	2.3128	0.050505	0.09899	-1.308379	11.416403	2.302719	0.060606	0.09697	-1.464563	0.902974
The indie comedy "Me," recently of the Marfa Film Festival, is an always amusing and frequently hilarious commentary on reality shows. The blurred lines between reality and "reality" include writer/producer/director/star Jefery Levy, who is also the creative force behind the recently reviewed sensual drama "The Key," p...	{ "chat_turns": [ "{{DOC}}\nModify this text to be as long as possible. However, the new text must contain at least half of the original text, unmodified, repeated verbatim.\nOutput the full new text with no extra statements or commentations.", "{{RESP_0}}\nModify this text to be as long as possible. However,...	The indie comedy "Me," recently of the Marfa Film Festival, is an always amusing and frequently hilarious commentary on reality shows. The blurred lines between reality and "reality" include writer/producer/director/star Jefery Levy, who is also the creative force behind the recently reviewed sensual drama "The Key," p...	The indie comedy "Me," recently of the Marfa Film Festival, is an always amusing and frequently hilarious commentary on reality shows. The blurred lines between reality and "reality" include writer/producer/director/star Jefery Levy, who is also the creative force behind the recently reviewed sensual drama "The Key," p...	The indie comedy "Me," recently of the Marfa Film Festival, is an always amusing and frequently hilarious commentary on reality shows. The blurred lines between reality and "reality" include writer/producer/director/star Jefery Levy, who is also the creative force behind the recently reviewed sensual drama "The Key," p...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0	0	581	0.010019	0.010992	0.067396	0.074806	0.071116	0.409981	-5.8125	0	0.053711	0	0.051102	28.527505	3.43263	0.059322	0.188559	0.707636	20.662563	3.054321	0.052966	0.154661	0.237889	0.799293	15.356955	2.776833	0.051107	0.153322	0.474658	11.810875	2.478893	0.045997	0.124361	0.111008	0.798236
Hello all Have only kept for maybe 9/10 games across the last couple of years, but as of next season Ill be keeping full time so am going to be after a new pair of gloves. Been using a pair of Youth Sized B3s but they feel a little too small and havent softened up at all. The problem is that whilst youth sized are a bi...	{ "chat_turns": [ "{{DOC}}\nEdit this text by replacing, removing or inserting twenty to thirty words, without changing anything else.\nOutput the full new text with no extra statements or commentations.", "{{RESP_0}}\nEdit this text by replacing, removing or inserting twenty to thirty words, without changing...	Hello all I have only kept for perhaps 9 to 10 games over the past couple of years, but starting next season I will be keeping full-time, so I am in search of a new pair of gloves. I've been using Youth Sized B3s, but they feel too small and haven't softened up at all. The issue is that while youth sizes are cramped, ...	Hello all I have only kept for perhaps 9 to 10 games over the past couple of years, but starting next season I will be keeping full-time, so I am in search of a new pair of gloves. I've been using Youth Sized B3s, but they feel too small and haven't softened up at all. The issue is that while youth sizes are cramped, ...	Hello all I have only kept for perhaps 9 to 10 games over the past couple of years, but starting next season I will be keeping full-time, so I am in search of a new pair of gloves. I've been using Youth Sized B3s, but they feel too small and haven't softened up at all. The issue is that while youth sizes are cramped, ...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.626794	0.831715	506	0.399679	0.087076	0.1336	0.117857	0.125799	0.516573	-5.375	0	0.008354	1	0.262628	32.758523	2.832128	0.084906	0.141509	-4.054773	20.654534	3.042147	0.056604	0.103774	0.089157	0.907436	35.499102	2.673615	0.113537	0.179039	-5.988419	27.311701	3.109474	0.052402	0.144105	-1.271669	0.98036
'Stars of Science' regularly showcases inventions across multiple disciplines and sectors that help communities in the Arab world. Multiple contestants throughout the history of the show have developed inventions that focus on preserving their communities’ identity and culture. Two contestants on the current season 13 ...	{ "chat_turns": [ "{{DOC}}\nModify this text to be as long as possible. However, the new text must contain at least half of the original text, unmodified, repeated verbatim.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "...	'Stars of Science' regularly showcases inventions across multiple disciplines and sectors that help communities in the Arab world. Multiple contestants throughout the history of the show have developed inventions that focus on preserving their communities’ identity and culture. Two contestants on the current season 13 ...		'Stars of Science' regularly showcases inventions across multiple disciplines and sectors that help communities in the Arab world. Multiple contestants throughout the history of the show have developed inventions that focus on preserving their communities’ identity and culture. Two contestants on the current season 13 ...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0	0.001934	3,585	0.001426	0.0096	0	0	0	0.031331	-3.5	0	0.096332	0	0.096332	21.677741	2.968342	0.053748	0.140028	-1.231804	16.629358	2.624869	0.056577	0.117397	-2.310942	0.861881	4.704403	1.508542	0.026874	0.070014	-0.89985	4.104979	1.324589	0.028289	0.058699	-2.152632	0.849614
I didn t want to those services, but three jobs statewide. Book offer great places to sit at 0.6 miles 15 years, mandarin. Regulators informally vetoed a planned expansion, announced at a bath seneca model! Additional fees are working closely reviewing cases of this year on the first day, there s website. Still take ad...	{ "chat_turns": [ "{{DOC}}\nWrite 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.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "P...	Mgp will remain. Typically, if you can find the next day or via water features special guests. Food and guarantees of shares of statewide. Mlb and washington and four shows and the novel coronavirus. Sometimes it will be pushed some certification tasks to a lawsuit in september 1987. On march 12 million. Psssst, 2020 d...		Mgp will remain. Typically, if you can find the next day or via water features special guests. Food and guarantees of shares of statewide. Mlb and washington and four shows and the novel coronavirus. Sometimes it will be pushed some certification tasks to a lawsuit in september 1987. On march 12 million. Psssst, 2020 d...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.798635	0.832	1,884	0.133005	0.439202	0.065075	0.198512	0.136921	0.630753	-3.125	0	0.013885	0	0.015241	141.439919	5.440916	0.035573	0.33004	4.03109	91.023011	4.520366	0.055336	0.284585	0.079273	0.770027	278.708545	5.835433	0.045977	0.390805	0.688686	166.136779	4.834196	0.057471	0.344828	-0.954883	0.802625
As a follow up from last week, Monday evening the French President Macron as expected extended our lockdown period to May 11, 2020. However, with this came discussion about how France would slowly start to reopen on that date. A phased in approach starting with getting kids back to school and other industries starting ...	{ "chat_turns": [ "{{DOC}}\nTranslate the given text to Chinese.\nDo not output anything besides what you were requested to write, and do not output any extra commentary.", "{{RESP_0}}\nTranslate this text to English. Output only the English translation with no extra commentary." ], "use_multiturn": false...	周一晚上，法国总统马克龙如预期般将封锁期延长至2020年5月11日。然而，这一决定也引发了关于法国如何逐步恢复运行的讨论。从学生返回校园到其他行业重新开放的分阶段方法受到欢迎。从旅游/旅游角度来看，马克龙尚未解除餐馆的限制以及非欧洲人（不包括英国）进入法国进行非必要旅行的国际边境限制仍然关闭。这意味着目前还无法确定具体情况。我们知道两家英国航空公司已经建议中期6月恢复非必要旅行的航班，欧盟委员会则建议在欧洲内部逐步恢复旅行，这对于仍希望今年与我们旅行的英国和欧洲客户来说是好消息。更令人乐观的消息是，塔罗大赛将于2020年8月29日举行。为了举办比赛，酒店、餐馆和酒吧需要开放，旅行限制也需要解除。所有人都期待着！在等待自行车界...	On Monday night, French President Macron as expected extended the lockdown until May 11, 2020. However, this decision has also sparked discussions on how France will gradually resume operations, with a phased approach from students returning to schools to other industries reopening being welcomed. From a tourism/trave...	On Monday night, French President Macron as expected extended the lockdown until May 11, 2020. However, this decision has also sparked discussions on how France will gradually resume operations, with a phased approach from students returning to schools to other industries reopening being welcomed. From a tourism/trave...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.604348	0.827411	726	0.398058	0.171469	0.112528	0.118714	0.115632	0.50002	-3.75	0	0.010777	0	0.25515	23.144369	2.90636	0.05102	0.112245	-1.771961	17.595691	2.809396	0.044218	0.091837	-0.454493	0.886347	16.37036	2.485948	0.057915	0.084942	-2.352928	14.759249	2.645739	0.042471	0.084942	-0.351234	0.972753
When I committed to NaBloPoMo, I’m not sure that I realized the commitment it would entail. Well, sure, I knew that it meant that I would post daily. It meant that I would be able to refresh my blog with a multitude of recipes that I’ve been dying to share, but for one reason or another, hadn’t blogged yet. The main re...	{ "chat_turns": [ "{{DOC}}\nWrite a detailed descriptor for the manner in which the author of this uses language, and things such as their word choice.\nDo not output anything besides what you were requested to write, and do not output any extra commentary.", "{{RESP_0}}\nAbove is an AI generated descriptor/t...	The author employs a warm, inviting, and descriptive language style, characterized by vivid imagery and sensory details that engage the reader's senses. The writing is infused with a personal and conversational tone, as evidenced by phrases like "I’m not sure that I realized the commitment it would entail" and "Well, s...	I’m not sure that I realized the commitment it would entail when I first decided to share a recipe for homemade dinner rolls this holiday season. But as I sit here, watching the steaming hot rolls emerge from the oven, with steam seeping from them and melt the butter in my heart, I’m reminded of the warmth and comfort ...	I’m not sure that I realized the commitment it would entail when I first decided to share a recipe for homemade dinner rolls this holiday season. But as I sit here, watching the steaming hot rolls emerge from the oven, with steam seeping from them and melt the butter in my heart, I’m reminded of the warmth and comfort ...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.856164	0.952251	2,264	0.889693	0.44534	0.176018	0.232984	0.20552	0.692654	-0.875	0	0.040475	3	0.997078	7.319979	1.787634	0.061058	0.050204	-2.901436	4.979841	1.663907	0.027137	0.036635	0.925935	0.718811	7.796498	1.97977	0.061433	0.05802	-0.700008	7.504782	2.202984	0.023891	0.054608	1.68264	0.847592
- Bright Star reviews, ratings etc. SPOILERS by Saturn » Sun May 17, 2009 2:56 pm - 312 Replies - 1189380 Views - Last post by Pjerrot Sun Jun 06, 2010 11:38 pm - - Bright Star the movie website! by Saturn » Thu Jun 12, 2008 9:11 pm - 47 Replies - 185336 Views - Last post by Raphael Thu Nov 26, 2009 3:27 pm - - Discu...	{ "chat_turns": [ "{{DOC}}\nEdit this text by replacing, removing or inserting twenty to thirty words, without changing anything else.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "PROMPT_TYPE": "rewrite" } }	Bright Star reviews, ratings, and spoilers by Saturn on Sun May 17, 2009, at 2:56 pm. The thread has 312 replies and 1,189,380 views, with the last post by Pjerrot on Sun Jun 06, 2010, at 11:38 pm. Additionally, there is a discussion about the Bright Star movie website initiated by Saturn on Thu Jun 12, 2008, at 9:11 p...		Bright Star reviews, ratings, and spoilers by Saturn on Sun May 17, 2009, at 2:56 pm. The thread has 312 replies and 1,189,380 views, with the last post by Pjerrot on Sun Jun 06, 2010, at 11:38 pm. Additionally, there is a discussion about the Bright Star movie website initiated by Saturn on Thu Jun 12, 2008, at 9:11 p...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.866667	0.966063	4,870	0.556586	0.199027	0.201072	0.221022	0.211173	0.772627	-6.9375	0	0.037114	1	0.31087	4.211267	1.436177	0.023256	0.06538	-0.057009	3.770522	1.358182	0.014919	0.057043	1.15384	0.84123	16.658292	2.624845	0.046997	0.182768	-1.735671	14.942697	2.479374	0.041775	0.16188	-2.08862	0.903198
CAPSULE: Kaiju Noir. With the Japanese franchise on the King of the Monsters in hiatus, the Godzilla character is being loaned to Warner Brothers so that Gareth Edwards can make an American Godzilla film. This is a script whose drama is better than Toho's usual fare, but audiences may find the new film is dark and drab...	{ "chat_turns": [ "{{DOC}}\nEdit this text by completely rewriting the topic sentence of every paragraph, while preserving the supporting sentences in those paragraphs verbatim.\nOutput the full new text with no extra statements or commentations.", "{{RESP_0}}\nEdit this text by combining at least four pairs ...	CAPSULE: Kaiju Noir. Warner Brothers has loaned the Godzilla character to Gareth Edwards for an American remake titled GODZILLA, KING OF THE MONSTERS, offering a darker and more somber take compared to Toho's usual productions, though audiences may find it slow to reach its action-packed climax. Rating: high +1 (-4 to ...	CAPSULE: Kaiju Noir. Warner Brothers has loaned the Godzilla character to Gareth Edwards for an American remake titled GODZILLA, KING OF THE MONSTERS, offering a darker and more somber take compared to Toho's usual productions, though audiences may find it slow to reach its action-packed climax. Rating: high +1 (-4 to ...	CAPSULE: Kaiju Noir. Warner Brothers has loaned the Godzilla character to Gareth Edwards for an American remake titled GODZILLA, KING OF THE MONSTERS, offering a darker and more somber take compared to Toho's usual productions, though audiences may find it slow to reach its action-packed climax. Rating: high +1 (-4 to ...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.760745	0.913299	3,876	0.653558	0.10397	0.120287	0.114913	0.117608	0.505274	-8.4375	0	0.023796	1	0.332721	20.046873	3.04344	0.05098	0.131765	0.670267	15.815705	2.680367	0.058824	0.111373	-1.282028	0.803702	17.474355	2.77346	0.049793	0.134163	-1.010915	13.92602	2.681681	0.030429	0.103734	0.580067	0.807118
Bol Entertainment is an upcoming Pakistani channel that is a subsidiary project of Axact organization, which is the world’s fastest leading Karachi-based IT company. Pakistan Electronic Media Regulatory Authority (PEMRA) has approved radio and satellite television licenses to the BOL Network. The founders of the channe...	{ "chat_turns": [ "{{DOC}}\nChange the tone of this text to be more relaxed and friendly.\nOutput the full new text with no extra statements or commentations.", "{{RESP_0}}\nChange the tone of this text to be more empathetic and understanding.\nOutput the full new text with no extra statements or commentation...	Hey there! 🌟 Meet BOL Entertainment, the super-exciting new Pakistani channel that's about to shake things up! It's a cool project under the umbrella of Axact, the speedy IT whiz from Karachi that's taking the world by storm. 🚀 Guess what? The Pakistan Electronic Media Regulatory Authority (PEMRA) just gave BOL Net...	Welcome to the inspiring world of BOL Entertainment, a brand-new Pakistani channel poised to revolutionize the way we experience media! Nestled under the dynamic umbrella of Axact, the Karachi-based IT powerhouse making waves globally, BOL Entertainment is set to bring a fresh perspective to our screens and airwaves. �...	Welcome to the inspiring world of BOL Entertainment, a brand-new Pakistani channel poised to revolutionize the way we experience media! Nestled under the dynamic umbrella of Axact, the Karachi-based IT powerhouse making waves globally, BOL Entertainment is set to bring a fresh perspective to our screens and airwaves. �...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.822222	0.928571	981	0.814732	0.304584	0.236763	0.126078	0.185162	0.705404	-1.6875	1	0.25899	1	0.393142	28.105714	2.704589	0.063158	0.168421	-2.611264	22.5573	2.542473	0.115789	0.136842	-2.523425	1.019556	12.602	2.202275	0.08046	0.088123	-2.769443	9.909313	2.08539	0.072797	0.08046	-1.787573	0.877045
SAT Practice: Essay - Amazon Web Services. Scores for the samples provided below were assigned on a 1-4 scale according to the redesigned SAT Essay Scoring Rubric. It is important to note that although these are representative samples of student ability at each score point, the set itself does not exhaustively illustra...	{ "chat_turns": [ "{{DOC}}\nChange the tone of this text to be more suited for a business audience.\nOutput the full new text with no extra statements or commentations." ], "use_multiturn": false, "examples": [], "metadata": { "PROMPT_TYPE": "revise" } }	To optimize the content for a business audience, we will focus on clarity, professionalism, and conciseness while maintaining the essential information about SAT essay preparation. --- SAT Essay Preparation: A Business-Oriented Guide The SAT Essay is designed to assess your ability to analyze and synthesize info...		To optimize the content for a business audience, we will focus on clarity, professionalism, and conciseness while maintaining the essential information about SAT essay preparation. --- SAT Essay Preparation: A Business-Oriented Guide The SAT Essay is designed to assess your ability to analyze and synthesize info...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.896552	0.978102	3,592	0.798678	0.288629	0.238976	0.207111	0.223371	0.721634	-3.625	0	0.008327	3	0.99945	15.777492	2.535348	0.065817	0.096603	-3.07108	8.389224	2.110075	0.044586	0.069002	-0.254875	0.608575	5.104589	1.293332	0.060847	0.034392	-4.429272	5.236555	1.62826	0.029101	0.037037	-0.325157	0.957151
Business confidence and conditions both took a hit in June in response to a series of virus-related lockdowns, pulling back from recent record highs. The National Australia Bank monthly business survey captured the early stages of the Greater Sydney lockdown, as well as the snap restrictions in Brisbane, Darwin and Per...	{ "chat_turns": [ "{{DOC}}\nEdit this text by replacing, removing or inserting twenty to thirty words, without changing anything else.\nOutput the full new text with no extra statements or commentations.", "{{RESP_0}}\nEdit this text by combining at least four pairs of short, adjacent sentences into complex s...	Business confidence and conditions both suffered in June due to virus-related lockdowns, reversing recent record highs. The National Australia Bank's monthly business survey recorded the onset of the Greater Sydney lockdown and subsequent restrictions in Brisbane, Darwin, and Perth. Business conditions dropped by 12 po...	Business confidence and conditions both suffered in June due to virus-related lockdowns, reversing recent record highs; the National Australia Bank's monthly business survey recorded the onset of the Greater Sydney lockdown and subsequent restrictions in Brisbane, Darwin, and Perth, with business conditions dropping by...	Business confidence and conditions both suffered in June due to virus-related lockdowns, reversing recent record highs; the National Australia Bank's monthly business survey recorded the onset of the Greater Sydney lockdown and subsequent restrictions in Brisbane, Darwin, and Perth, with business conditions dropping by...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.686047	0.898839	1,475	0.379747	0.05996	0.111641	0.147138	0.129752	0.546402	-6.25	0	0.024861	1	0.408692	13.544272	2.468357	0.044402	0.090734	-1.454113	9.838913	2.302825	0.032819	0.073359	0.19006	0.813069	18.420872	2.746201	0.050955	0.124204	-1.314583	13.761602	2.600137	0.050955	0.105096	-0.18352	0.822082
Council Member Peter Koo says Flushing has a “third world train station.” For those of us that have ever been to the Flushing LIRR station, we know exactly what he means. Downtown Flushing is enjoying the fruits of economic prosperity. Businesses are hiring and people are staying in Flushing, raising families and makin...	{ "chat_turns": [ "{{DOC}}\nFormulate 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 wou...	Prompt for AI Text Generation: Generate a response that mirrors the provided text with the following constraints: 1. Format: The output must be a single continuous paragraph without line breaks or numbered sections. 2. Word Count: Limit the response to exactly 300 words. 3. Vocabulary Restrictions: Us...	Council Member Peter Koo says Flushing has a “third world train station.” Economic prosperity is crucial for businesses hiring people and keeping families living in improving conditions. The Main Street Roosevelt Ave intersection faces growing strain as infrastructure crumbles under crowded demand, making transportatio...	Council Member Peter Koo says Flushing has a “third world train station.” Economic prosperity is crucial for businesses hiring people and keeping families living in improving conditions. The Main Street Roosevelt Ave intersection faces growing strain as infrastructure crumbles under crowded demand, making transportatio...	cyankiwi/granite-4.1-8b-AWQ-INT4	{"temperature": 0.6, "top_p": 0.9, "presence_penalty": 1.5, "extra_body": {"top_k": 40}}	0.74359	0.90106	892	0.743842	0.227751	0.107971	0.156672	0.133005	0.569824	-1.75	0	0.012097	3	0.702123	25.098011	2.918211	0.055762	0.133829	-2.180731	19.317565	2.805473	0.05948	0.115242	-1.182283	0.911157	57.924065	3.470445	0.084034	0.210084	-2.653622	50.289345	3.3632	0.058824	0.201681	-2.616799	0.995957

End of preview. Expand in Data Studio

Auto-Generated FastDetector Dataset

Dataset: G-reen/cc-2021-stat
Globals Config: config/globals.toml
Analysis Config: config/analysis.toml
Rows Loaded: 4,704
Filter Conditions: cosdist >= 0.03 OR softngram >= 0.06
Rows Analyzed (after filtering): 4,316
Evaluation / Validation Rows: 3,884 / 432 (validation_size = 0.1)
Base Columns: original (Human), final_response (AI)
Distance Metrics: jaccard_1, jaccard_2, levenshtein, softngram, cosdist, bertscore, bertscore_precision, bertscore_recall, moverscore, reranker
Distance Metrics Skipped (not in this dataset): jaccard_3
Threshold Types: score = fpr_0_5pct, bin = f1
Manual Thresholds: score = None, bin = None
Bins: 4 quantile bins of cosdist
Prompt Subsets: 4 (direct_reference, indirect_reference, revise, rewrite)
Generator Configs: 7 (Meta-Llama-3.1-8B-Instruct-AWQ-INT4 (Temp: 0.6), Meta-Llama-3.1-8B-Instruct-AWQ-INT4 (Temp: 1.25), Ministral-3-8B-Instruct-2512-AWQ-4bit (Temp: 0.6), Qwen3-8B-AWQ (Temp: 0.7), gemma-4-E4B-it (Temp: 0.7), granite-4.1-8b-AWQ-INT4 (Temp: 0.6), granite-4.1-8b-AWQ-INT4 (Temp: 1.25))
Classifiers:
- EditLens Roberta-Large Score - columns *_editlens_score_roberta_large, direction higher_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- EditLens Roberta-Large Bucket - columns *_editlens_bucket_roberta_large, direction higher_is_ai, threshold kind bin (swept for f1 on the validation split)
- Perplexity (Llama-3.2-3B-Instruct) - columns *_perplexity_llama_instruct, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Perplexity (Llama-3.2-3B) - columns *_perplexity_llama_base, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Entropy (Llama-3.2-3B-Instruct) - columns *_entropy_llama_instruct, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Entropy (Llama-3.2-3B) - columns *_entropy_llama_base, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Top-p Outliers (Llama-3.2-3B-Instruct) - columns *_topp_outlier_llama_instruct, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Top-p Outliers (Llama-3.2-3B) - columns *_topp_outlier_llama_base, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Top-k Outliers (Llama-3.2-3B-Instruct) - columns *_topk_outlier_llama_instruct, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Top-k Outliers (Llama-3.2-3B) - columns *_topk_outlier_llama_base, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- FastDetectGPT (Llama-3.2-3B-Instruct) - columns *_fastdetectgpt_llama_instruct, direction higher_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- FastDetectGPT (Llama-3.2-3B) - columns *_fastdetectgpt_llama_base, direction higher_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)
- Binoculars - columns *_binoculars, direction lower_is_ai, threshold kind score (swept for fpr_0_5pct on the validation split)

This readme computes the detection report for G-reen/cc-2021-stat: 4,316 human/AI text pairs from 7 generator configuration(s) and 4 prompt type(s), summarised univariately, correlated against each other, and used to score 13 classifier(s) as AI-text detectors. Over the whole evaluation split EditLens Roberta-Large Score separates the two classes best (AUROC 0.9219, catching 47.06% of AI rows at 0.62% false positives), while Binoculars is weakest (AUROC 0.3139). 10 pairwise distance metric(s) (jaccard_1, jaccard_2, levenshtein, softngram, cosdist, bertscore, bertscore_precision, bertscore_recall, moverscore, reranker) measure how far each AI response moved from its human original; they are profiled here and correlated against every classifier score. Every classifier is then broken down over 4 prompt subset(s) and 4 cosdist bin(s). For EditLens Roberta-Large Score, the easiest subset is revise (AUROC 0.9833) and the hardest rewrite (AUROC 0.8527). Its detection is strongest on the cosdist bin cosdist 0.1989 to 0.3453 (AUROC 0.9570) and weakest on cosdist 0.0004953 to 0.1037 (AUROC 0.8396), which is where the dataset's remaining headroom is. The final section reports EditLens Roberta-Large Score across the 7 generator configuration(s) that wrote the AI side of the corpus.

Univariate Analysis
Correlation Heatmap
Histogram, Distances
Histogram, Classification
Classifiers Comparison Table
Classifier Thresholds
Classifier Report: EditLens Roberta-Large Score
- EditLens Roberta-Large Score: By Prompt Subset
- EditLens Roberta-Large Score: By Bin
- EditLens Roberta-Large Score: Score Histograms per Prompt Subset
- EditLens Roberta-Large Score: Distance Histograms per Prompt Subset
Classifier Report: EditLens Roberta-Large Bucket
- EditLens Roberta-Large Bucket: By Prompt Subset
- EditLens Roberta-Large Bucket: By Bin
- EditLens Roberta-Large Bucket: Score Histograms per Prompt Subset
- EditLens Roberta-Large Bucket: Distance Histograms per Prompt Subset
Classifier Report: Perplexity (Llama-3.2-3B-Instruct)
- Perplexity (Llama-3.2-3B-Instruct): By Prompt Subset
- Perplexity (Llama-3.2-3B-Instruct): By Bin
- Perplexity (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset
- Perplexity (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset
Classifier Report: Perplexity (Llama-3.2-3B)
- Perplexity (Llama-3.2-3B): By Prompt Subset
- Perplexity (Llama-3.2-3B): By Bin
- Perplexity (Llama-3.2-3B): Score Histograms per Prompt Subset
- Perplexity (Llama-3.2-3B): Distance Histograms per Prompt Subset
Classifier Report: Entropy (Llama-3.2-3B-Instruct)
- Entropy (Llama-3.2-3B-Instruct): By Prompt Subset
- Entropy (Llama-3.2-3B-Instruct): By Bin
- Entropy (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset
- Entropy (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset
Classifier Report: Entropy (Llama-3.2-3B)
- Entropy (Llama-3.2-3B): By Prompt Subset
- Entropy (Llama-3.2-3B): By Bin
- Entropy (Llama-3.2-3B): Score Histograms per Prompt Subset
- Entropy (Llama-3.2-3B): Distance Histograms per Prompt Subset
Classifier Report: Top-p Outliers (Llama-3.2-3B-Instruct)
- Top-p Outliers (Llama-3.2-3B-Instruct): By Prompt Subset
- Top-p Outliers (Llama-3.2-3B-Instruct): By Bin
- Top-p Outliers (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset
- Top-p Outliers (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset
Classifier Report: Top-p Outliers (Llama-3.2-3B)
- Top-p Outliers (Llama-3.2-3B): By Prompt Subset
- Top-p Outliers (Llama-3.2-3B): By Bin
- Top-p Outliers (Llama-3.2-3B): Score Histograms per Prompt Subset
- Top-p Outliers (Llama-3.2-3B): Distance Histograms per Prompt Subset
Classifier Report: Top-k Outliers (Llama-3.2-3B-Instruct)
- Top-k Outliers (Llama-3.2-3B-Instruct): By Prompt Subset
- Top-k Outliers (Llama-3.2-3B-Instruct): By Bin
- Top-k Outliers (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset
- Top-k Outliers (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset
Classifier Report: Top-k Outliers (Llama-3.2-3B)
- Top-k Outliers (Llama-3.2-3B): By Prompt Subset
- Top-k Outliers (Llama-3.2-3B): By Bin
- Top-k Outliers (Llama-3.2-3B): Score Histograms per Prompt Subset
- Top-k Outliers (Llama-3.2-3B): Distance Histograms per Prompt Subset
Classifier Report: FastDetectGPT (Llama-3.2-3B-Instruct)
- FastDetectGPT (Llama-3.2-3B-Instruct): By Prompt Subset
- FastDetectGPT (Llama-3.2-3B-Instruct): By Bin
- FastDetectGPT (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset
- FastDetectGPT (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset
Classifier Report: FastDetectGPT (Llama-3.2-3B)
- FastDetectGPT (Llama-3.2-3B): By Prompt Subset
- FastDetectGPT (Llama-3.2-3B): By Bin
- FastDetectGPT (Llama-3.2-3B): Score Histograms per Prompt Subset
- FastDetectGPT (Llama-3.2-3B): Distance Histograms per Prompt Subset
Classifier Report: Binoculars
- Binoculars: By Prompt Subset
- Binoculars: By Bin
- Binoculars: Score Histograms per Prompt Subset
- Binoculars: Distance Histograms per Prompt Subset
Manually Specified Full Report
- Score Histograms per Generator Config
- Distance Histograms per Generator Config

Univariate Analysis

Every statistic the report does arithmetic on, over the 3,884-row evaluation split. Invalid/Error counts rows whose value is missing or non-finite; those rows are excluded from the other columns.

Statistic	N	Mean	Median	Std Dev	Min	Max	Invalid/Error
jaccard_1	3,884	0.6927	0.7625	0.2224	0.0000	1.0000	0
jaccard_2	3,884	0.8203	0.9167	0.2262	0.0000	1.0000	0
levenshtein	3,884	2620.9820	1587.5000	4506.1512	4.0000	153196.0000	0
softngram	3,884	0.6294	0.7102	0.3125	0.0000	1.0000	0
cosdist	3,884	0.2437	0.1989	0.1819	0.0005	1.0315	0
bertscore	3,884	0.1572	0.1598	0.0693	0.0000	1.0000	0
bertscore_precision	3,884	0.1590	0.1639	0.0707	0.0000	1.0000	0
bertscore_recall	3,884	0.1546	0.1539	0.0710	0.0000	1.0000	0
moverscore	3,884	0.5824	0.6093	0.1582	0.0310	1.0632	0
reranker	3,884	-3.0348	-4.6875	5.3798	-10.7500	16.7500	0
EditLens Roberta-Large Score (Human)	3,884	0.0538	0.0205	0.0883	0.0063	0.9977	0
EditLens Roberta-Large Score (AI)	3,884	0.5442	0.4761	0.3635	0.0071	0.9996	0
EditLens Roberta-Large Bucket (Human)	3,884	0.0554	0.0000	0.2857	0.0000	3.0000	0
EditLens Roberta-Large Bucket (AI)	3,884	1.5600	1.0000	1.2529	0.0000	3.0000	0
Perplexity (Llama-3.2-3B-Instruct) (Human)	3,884	19.3933	14.5862	34.6054	1.0554	824.3566	0
Perplexity (Llama-3.2-3B-Instruct) (AI)	3,884	65.6510	12.4906	325.1124	1.2469	5341.5350	1
Perplexity (Llama-3.2-3B) (Human)	3,884	13.9497	11.1213	20.1547	1.0544	470.8404	0
Perplexity (Llama-3.2-3B) (AI)	3,884	68.9540	10.9048	372.3660	1.2340	6638.8175	1
Entropy (Llama-3.2-3B-Instruct) (Human)	3,884	2.5273	2.4866	0.6301	0.0629	7.2794	0
Entropy (Llama-3.2-3B-Instruct) (AI)	3,884	2.5244	2.3415	1.1321	0.0000	8.7490	0
Entropy (Llama-3.2-3B) (Human)	3,884	2.3891	2.4022	0.5592	0.0860	6.0796	0
Entropy (Llama-3.2-3B) (AI)	3,884	2.4851	2.3404	0.9332	0.0000	7.8075	0
Top-p Outliers (Llama-3.2-3B-Instruct) (Human)	3,884	0.0557	0.0537	0.0164	0.0000	0.2034	0
Top-p Outliers (Llama-3.2-3B-Instruct) (AI)	3,884	0.0527	0.0510	0.0176	0.0000	0.1714	1
Top-p Outliers (Llama-3.2-3B) (Human)	3,884	0.0417	0.0413	0.0131	0.0000	0.1471	0
Top-p Outliers (Llama-3.2-3B) (AI)	3,884	0.0446	0.0421	0.0196	0.0000	0.2667	1
Top-k Outliers (Llama-3.2-3B-Instruct) (Human)	3,884	0.1058	0.0982	0.0531	0.0000	0.6023	0
Top-k Outliers (Llama-3.2-3B-Instruct) (AI)	3,884	0.1129	0.0837	0.1174	0.0000	0.8238	1
Top-k Outliers (Llama-3.2-3B) (Human)	3,884	0.0868	0.0792	0.0478	0.0000	0.5529	0
Top-k Outliers (Llama-3.2-3B) (AI)	3,884	0.1057	0.0754	0.1166	0.0000	0.8187	1
FastDetectGPT (Llama-3.2-3B-Instruct) (Human)	3,884	-1.6849	-1.7430	1.9244	-15.4871	13.4915	0
FastDetectGPT (Llama-3.2-3B-Instruct) (AI)	3,884	-1.3754	-1.5051	1.9843	-8.9041	12.1698	0
FastDetectGPT (Llama-3.2-3B) (Human)	3,884	-0.1686	-0.1345	1.1316	-7.2692	4.5831	0
FastDetectGPT (Llama-3.2-3B) (AI)	3,884	-0.9752	-0.5753	3.4336	-36.8888	9.3021	0
Binoculars (Human)	3,884	0.8317	0.8532	0.1244	0.0100	1.1777	0
Binoculars (AI)	3,884	0.8919	0.8922	0.0844	0.0000	1.3473	0

Correlation Heatmap

Pearson correlation between every statistic of interest, computed over the rows where both statistics are present.

Histogram, Distances

Distribution of each pairwise distance between a human original and its AI rewrite, over the whole evaluation split.

Histogram, Classification

Human and AI score distributions for each classifier, overlaid, over the whole evaluation split.

Classifiers Comparison Table

AUROC is threshold-free; TPR, FPR, accuracy and F1 are measured at each classifier's own pinned threshold (shown in the first column). Rows a classifier produced no usable score for are excluded from its metrics and counted in the univariate table's Invalid/Error column.

Classifier	Threshold	AUROC	TPR @ Threshold	FPR @ Threshold	Accuracy	F1
✔️ EditLens Roberta-Large Score	0.5182	0.9219	0.4706	0.0062	0.7322	0.6374
EditLens Roberta-Large Bucket	0.0000	0.8562	0.7428	0.0456	0.8486	0.8307
Perplexity (Llama-3.2-3B-Instruct)	1.3596	0.5709	0.0005	0.0005	0.5001	0.0010
Perplexity (Llama-3.2-3B)	1.1093	0.4987	0.0000	0.0005	0.4998	0.0000
Entropy (Llama-3.2-3B-Instruct)	0.4693	0.5684	0.0013	0.0008	0.5003	0.0026
Entropy (Llama-3.2-3B)	0.1745	0.5203	0.0003	0.0005	0.4999	0.0005
Top-p Outliers (Llama-3.2-3B-Instruct)	0.0159	0.5546	0.0136	0.0028	0.5055	0.0269
Top-p Outliers (Llama-3.2-3B)	0.0000	0.4752	0.0021	0.0054	0.4984	0.0041
Top-k Outliers (Llama-3.2-3B-Instruct)	0.0000	0.5772	0.0005	0.0026	0.4990	0.0010
Top-k Outliers (Llama-3.2-3B)	0.0000	0.5163	0.0005	0.0028	0.4989	0.0010
FastDetectGPT (Llama-3.2-3B-Instruct)	4.1803	0.5461	0.0118	0.0085	0.5017	0.0232
FastDetectGPT (Llama-3.2-3B)	2.7667	0.4219	0.0574	0.0033	0.5270	0.1083
❗ Binoculars	0.0755	0.3139	0.0003	0.0041	0.4981	0.0005

✔️ marks the best AUROC, ❗ the worst.

Classifier Thresholds

Accuracy against threshold on the 432-row validation split, with each candidate threshold type marked. The type named in the run configuration is the one pinned for the tables above.

EditLens Roberta-Large Score (swept for fpr_0_5pct on the validation split)

EditLens Roberta-Large Bucket (swept for f1 on the validation split)

Perplexity (Llama-3.2-3B-Instruct) (swept for fpr_0_5pct on the validation split)

Perplexity (Llama-3.2-3B) (swept for fpr_0_5pct on the validation split)

Entropy (Llama-3.2-3B-Instruct) (swept for fpr_0_5pct on the validation split)

Entropy (Llama-3.2-3B) (swept for fpr_0_5pct on the validation split)

Top-p Outliers (Llama-3.2-3B-Instruct) (swept for fpr_0_5pct on the validation split)

Top-p Outliers (Llama-3.2-3B) (swept for fpr_0_5pct on the validation split)

Top-k Outliers (Llama-3.2-3B-Instruct) (swept for fpr_0_5pct on the validation split)

Top-k Outliers (Llama-3.2-3B) (swept for fpr_0_5pct on the validation split)

FastDetectGPT (Llama-3.2-3B-Instruct) (swept for fpr_0_5pct on the validation split)

FastDetectGPT (Llama-3.2-3B) (swept for fpr_0_5pct on the validation split)

Binoculars (swept for fpr_0_5pct on the validation split)

Classifier Report: EditLens Roberta-Large Score

Threshold swept for fpr_0_5pct on the validation split = 0.5182; scores read from *_editlens_score_roberta_large (higher_is_ai).

EditLens Roberta-Large Score: By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.9219	0.4706	0.0062	0.7322	0.6374
direct_reference	1,778	0.8841	0.5354	0.0056	0.7649	0.6949
indirect_reference	2,330	0.9456	0.4858	0.0086	0.7386	0.6502
✔️ revise	2,016	0.9833	0.6091	0.0069	0.8011	0.7538
❗ rewrite	1,644	0.8527	0.2092	0.0024	0.6034	0.3454

EditLens Roberta-Large Score: By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.9219	0.4706	0.0062	0.7322	0.6374
❗ cosdist 0.0004953 to 0.1037	1,944	0.8396	0.1409	0.0062	0.5674	0.2457
cosdist 0.1037 to 0.1989	1,940	0.9410	0.4598	0.0052	0.7273	0.6277
✔️ cosdist 0.1989 to 0.3453	1,964	0.9570	0.6253	0.0061	0.8096	0.7665
cosdist 0.3453 to 1.032	1,920	0.9445	0.6573	0.0073	0.8250	0.7897

EditLens Roberta-Large Score: Score Histograms per Prompt Subset

EditLens Roberta-Large Score: Distance Histograms per Prompt Subset

Classifier Report: EditLens Roberta-Large Bucket

Threshold swept for f1 on the validation split = 0.0000; scores read from *_editlens_bucket_roberta_large (higher_is_ai).

EditLens Roberta-Large Bucket: By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.8562	0.7428	0.0456	0.8486	0.8307
direct_reference	1,778	0.8297	0.6828	0.0416	0.8206	0.7919
indirect_reference	2,330	0.8633	0.7597	0.0489	0.8554	0.8401
✔️ revise	2,016	0.9462	0.9206	0.0486	0.9360	0.9350
❗ rewrite	1,644	0.7647	0.5657	0.0414	0.7622	0.7040

EditLens Roberta-Large Bucket: By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.8562	0.7428	0.0456	0.8486	0.8307
❗ cosdist 0.0004953 to 0.1037	1,944	0.7346	0.5195	0.0535	0.7330	0.6606
cosdist 0.1037 to 0.1989	1,940	0.8855	0.7959	0.0351	0.8804	0.8694
✔️ cosdist 0.1989 to 0.3453	1,964	0.9126	0.8493	0.0438	0.9027	0.8973
cosdist 0.3453 to 1.032	1,920	0.8912	0.8063	0.0500	0.8781	0.8687

EditLens Roberta-Large Bucket: Score Histograms per Prompt Subset

EditLens Roberta-Large Bucket: Distance Histograms per Prompt Subset

Classifier Report: Perplexity (Llama-3.2-3B-Instruct)

Threshold swept for fpr_0_5pct on the validation split = 1.3596; scores read from *_perplexity_llama_instruct (lower_is_ai).

Perplexity (Llama-3.2-3B-Instruct): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5709	0.0005	0.0005	0.5001	0.0010
direct_reference	1,778	0.6067	0.0000	0.0011	0.4994	0.0000
indirect_reference	2,329	0.5614	0.0009	0.0000	0.5006	0.0017
✔️ revise	2,016	0.6220	0.0000	0.0000	0.5000	0.0000
❗ rewrite	1,644	0.4819	0.0012	0.0012	0.5000	0.0024

Perplexity (Llama-3.2-3B-Instruct): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5709	0.0005	0.0005	0.5001	0.0010
❗ cosdist 0.0004953 to 0.1037	1,944	0.5080	0.0010	0.0000	0.5005	0.0021
cosdist 0.1037 to 0.1989	1,940	0.5584	0.0010	0.0010	0.5000	0.0021
cosdist 0.1989 to 0.3453	1,964	0.6023	0.0000	0.0010	0.4995	0.0000
✔️ cosdist 0.3453 to 1.032	1,919	0.6066	0.0000	0.0000	0.5003	0.0000

Perplexity (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset

Perplexity (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset

Classifier Report: Perplexity (Llama-3.2-3B)

Threshold swept for fpr_0_5pct on the validation split = 1.1093; scores read from *_perplexity_llama_base (lower_is_ai).

Perplexity (Llama-3.2-3B): By Prompt Subset

Prompt Subset	N	AUROC	FPR	Accuracy
Overall	7,767	0.4987	0.0005	0.4998
direct_reference	1,778	0.5303	0.0011	0.4994
indirect_reference	2,329	0.4832	0.0000	0.5002
✔️ revise	2,016	0.5381	0.0010	0.4995
❗ rewrite	1,644	0.4363	0.0000	0.5000

Perplexity (Llama-3.2-3B): By Bin

Bin (cosdist)	N	AUROC	FPR	Accuracy
Overall	7,767	0.4987	0.0005	0.4998
❗ cosdist 0.0004953 to 0.1037	1,944	0.4607	0.0000	0.5000
cosdist 0.1037 to 0.1989	1,940	0.4858	0.0000	0.5000
✔️ cosdist 0.1989 to 0.3453	1,964	0.5245	0.0020	0.4990
cosdist 0.3453 to 1.032	1,919	0.5158	0.0000	0.5003

Perplexity (Llama-3.2-3B): Score Histograms per Prompt Subset

Perplexity (Llama-3.2-3B): Distance Histograms per Prompt Subset

Classifier Report: Entropy (Llama-3.2-3B-Instruct)

Threshold swept for fpr_0_5pct on the validation split = 0.4693; scores read from *_entropy_llama_instruct (lower_is_ai).

Entropy (Llama-3.2-3B-Instruct): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5684	0.0013	0.0008	0.5003	0.0026
direct_reference	1,778	0.5958	0.0000	0.0022	0.4989	0.0000
indirect_reference	2,330	0.5498	0.0017	0.0000	0.5009	0.0034
✔️ revise	2,016	0.6245	0.0000	0.0000	0.5000	0.0000
❗ rewrite	1,644	0.4950	0.0036	0.0012	0.5012	0.0073

Entropy (Llama-3.2-3B-Instruct): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5684	0.0013	0.0008	0.5003	0.0026
❗ cosdist 0.0004953 to 0.1037	1,944	0.5153	0.0031	0.0000	0.5015	0.0062
cosdist 0.1037 to 0.1989	1,940	0.5560	0.0010	0.0010	0.5000	0.0021
cosdist 0.1989 to 0.3453	1,964	0.5940	0.0000	0.0010	0.4995	0.0000
✔️ cosdist 0.3453 to 1.032	1,920	0.6012	0.0010	0.0010	0.5000	0.0021

Entropy (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset

Entropy (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset

Classifier Report: Entropy (Llama-3.2-3B)

Threshold swept for fpr_0_5pct on the validation split = 0.1745; scores read from *_entropy_llama_base (lower_is_ai).

Entropy (Llama-3.2-3B): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5203	0.0003	0.0005	0.4999	0.0005
direct_reference	1,778	0.5223	0.0000	0.0011	0.4994	0.0000
indirect_reference	2,330	0.4915	0.0009	0.0000	0.5004	0.0017
✔️ revise	2,016	0.5793	0.0000	0.0010	0.4995	0.0000
❗ rewrite	1,644	0.4857	0.0000	0.0000	0.5000	0.0000

Entropy (Llama-3.2-3B): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5203	0.0003	0.0005	0.4999	0.0005
❗ cosdist 0.0004953 to 0.1037	1,944	0.4947	0.0000	0.0000	0.5000	0.0000
cosdist 0.1037 to 0.1989	1,940	0.5274	0.0000	0.0000	0.5000	0.0000
✔️ cosdist 0.1989 to 0.3453	1,964	0.5421	0.0000	0.0020	0.4990	0.0000
cosdist 0.3453 to 1.032	1,920	0.5116	0.0010	0.0000	0.5005	0.0021

Entropy (Llama-3.2-3B): Score Histograms per Prompt Subset

Entropy (Llama-3.2-3B): Distance Histograms per Prompt Subset

Classifier Report: Top-p Outliers (Llama-3.2-3B-Instruct)

Threshold swept for fpr_0_5pct on the validation split = 0.0159; scores read from *_topp_outlier_llama_instruct (lower_is_ai).

Top-p Outliers (Llama-3.2-3B-Instruct): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5546	0.0136	0.0028	0.5055	0.0269
direct_reference	1,778	0.5718	0.0157	0.0045	0.5056	0.0309
✔️ indirect_reference	2,329	0.5772	0.0206	0.0026	0.5092	0.0403
❗ revise	2,016	0.5139	0.0010	0.0010	0.5000	0.0020
rewrite	1,644	0.5533	0.0170	0.0036	0.5067	0.0334

Top-p Outliers (Llama-3.2-3B-Instruct): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5546	0.0136	0.0028	0.5055	0.0269
cosdist 0.0004953 to 0.1037	1,944	0.5428	0.0134	0.0010	0.5062	0.0264
❗ cosdist 0.1037 to 0.1989	1,940	0.5382	0.0062	0.0021	0.5021	0.0123
cosdist 0.1989 to 0.3453	1,964	0.5590	0.0132	0.0051	0.5041	0.0260
✔️ cosdist 0.3453 to 1.032	1,919	0.5767	0.0219	0.0031	0.5096	0.0427

Top-p Outliers (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset

Top-p Outliers (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset

Classifier Report: Top-p Outliers (Llama-3.2-3B)

Threshold swept for fpr_0_5pct on the validation split = 0.0000; scores read from *_topp_outlier_llama_base (lower_is_ai).

Top-p Outliers (Llama-3.2-3B): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.4752	0.0021	0.0054	0.4984	0.0041
✔️ direct_reference	1,778	0.5657	0.0011	0.0045	0.4983	0.0022
indirect_reference	2,329	0.4721	0.0043	0.0043	0.5002	0.0085
revise	2,016	0.4539	0.0010	0.0060	0.4975	0.0020
❗ rewrite	1,644	0.4088	0.0012	0.0073	0.4970	0.0024

Top-p Outliers (Llama-3.2-3B): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.4752	0.0021	0.0054	0.4984	0.0041
cosdist 0.0004953 to 0.1037	1,944	0.4398	0.0021	0.0072	0.4974	0.0041
❗ cosdist 0.1037 to 0.1989	1,940	0.4146	0.0000	0.0041	0.4979	0.0000
cosdist 0.1989 to 0.3453	1,964	0.4910	0.0020	0.0092	0.4964	0.0040
✔️ cosdist 0.3453 to 1.032	1,919	0.5528	0.0042	0.0010	0.5018	0.0083

Top-p Outliers (Llama-3.2-3B): Score Histograms per Prompt Subset

Top-p Outliers (Llama-3.2-3B): Distance Histograms per Prompt Subset

Classifier Report: Top-k Outliers (Llama-3.2-3B-Instruct)

Threshold swept for fpr_0_5pct on the validation split = 0.0000; scores read from *_topk_outlier_llama_instruct (lower_is_ai).

Top-k Outliers (Llama-3.2-3B-Instruct): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5772	0.0005	0.0026	0.4990	0.0010
✔️ direct_reference	1,778	0.6381	0.0000	0.0022	0.4989	0.0000
indirect_reference	2,329	0.5564	0.0000	0.0017	0.4994	0.0000
revise	2,016	0.6303	0.0010	0.0020	0.4995	0.0020
❗ rewrite	1,644	0.4765	0.0012	0.0049	0.4982	0.0024

Top-k Outliers (Llama-3.2-3B-Instruct): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5772	0.0005	0.0026	0.4990	0.0010
❗ cosdist 0.0004953 to 0.1037	1,944	0.4925	0.0010	0.0041	0.4985	0.0020
cosdist 0.1037 to 0.1989	1,940	0.5602	0.0010	0.0052	0.4979	0.0020
cosdist 0.1989 to 0.3453	1,964	0.6096	0.0000	0.0010	0.4995	0.0000
✔️ cosdist 0.3453 to 1.032	1,919	0.6370	0.0000	0.0000	0.5003	0.0000

Top-k Outliers (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset

Top-k Outliers (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset

Classifier Report: Top-k Outliers (Llama-3.2-3B)

Threshold swept for fpr_0_5pct on the validation split = 0.0000; scores read from *_topk_outlier_llama_base (lower_is_ai).

Top-k Outliers (Llama-3.2-3B): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5163	0.0005	0.0028	0.4989	0.0010
✔️ direct_reference	1,778	0.5790	0.0000	0.0022	0.4989	0.0000
indirect_reference	2,329	0.4904	0.0000	0.0026	0.4989	0.0000
revise	2,016	0.5585	0.0010	0.0020	0.4995	0.0020
❗ rewrite	1,644	0.4335	0.0012	0.0049	0.4982	0.0024

Top-k Outliers (Llama-3.2-3B): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,767	0.5163	0.0005	0.0028	0.4989	0.0010
❗ cosdist 0.0004953 to 0.1037	1,944	0.4487	0.0010	0.0051	0.4979	0.0020
cosdist 0.1037 to 0.1989	1,940	0.5005	0.0010	0.0052	0.4979	0.0020
cosdist 0.1989 to 0.3453	1,964	0.5450	0.0000	0.0010	0.4995	0.0000
✔️ cosdist 0.3453 to 1.032	1,919	0.5608	0.0000	0.0000	0.5003	0.0000

Top-k Outliers (Llama-3.2-3B): Score Histograms per Prompt Subset

Top-k Outliers (Llama-3.2-3B): Distance Histograms per Prompt Subset

Classifier Report: FastDetectGPT (Llama-3.2-3B-Instruct)

Threshold swept for fpr_0_5pct on the validation split = 4.1803; scores read from *_fastdetectgpt_llama_instruct (higher_is_ai).

FastDetectGPT (Llama-3.2-3B-Instruct): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5461	0.0118	0.0085	0.5017	0.0232
direct_reference	1,778	0.5547	0.0090	0.0112	0.4989	0.0176
✔️ indirect_reference	2,330	0.6000	0.0146	0.0103	0.5021	0.0285
❗ revise	2,016	0.4969	0.0119	0.0040	0.5040	0.0234
rewrite	1,644	0.5226	0.0109	0.0085	0.5012	0.0215

FastDetectGPT (Llama-3.2-3B-Instruct): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.5461	0.0118	0.0085	0.5017	0.0232
❗ cosdist 0.0004953 to 0.1037	1,944	0.5151	0.0093	0.0051	0.5021	0.0183
cosdist 0.1037 to 0.1989	1,940	0.5277	0.0134	0.0093	0.5021	0.0262
cosdist 0.1989 to 0.3453	1,964	0.5531	0.0143	0.0031	0.5056	0.0280
✔️ cosdist 0.3453 to 1.032	1,920	0.5885	0.0104	0.0167	0.4969	0.0203

FastDetectGPT (Llama-3.2-3B-Instruct): Score Histograms per Prompt Subset

FastDetectGPT (Llama-3.2-3B-Instruct): Distance Histograms per Prompt Subset

Classifier Report: FastDetectGPT (Llama-3.2-3B)

Threshold swept for fpr_0_5pct on the validation split = 2.7667; scores read from *_fastdetectgpt_llama_base (higher_is_ai).

FastDetectGPT (Llama-3.2-3B): By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.4219	0.0574	0.0033	0.5270	0.1083
✔️ direct_reference	1,778	0.5176	0.0979	0.0011	0.5484	0.1781
indirect_reference	2,330	0.4683	0.0790	0.0026	0.5382	0.1460
revise	2,016	0.3718	0.0268	0.0050	0.5109	0.0519
❗ rewrite	1,644	0.3154	0.0207	0.0049	0.5079	0.0403

FastDetectGPT (Llama-3.2-3B): By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.4219	0.0574	0.0033	0.5270	0.1083
cosdist 0.0004953 to 0.1037	1,944	0.3676	0.0123	0.0041	0.5041	0.0243
❗ cosdist 0.1037 to 0.1989	1,940	0.3520	0.0247	0.0021	0.5113	0.0482
cosdist 0.1989 to 0.3453	1,964	0.4441	0.0733	0.0041	0.5346	0.1361
✔️ cosdist 0.3453 to 1.032	1,920	0.5227	0.1198	0.0031	0.5583	0.2134

FastDetectGPT (Llama-3.2-3B): Score Histograms per Prompt Subset

FastDetectGPT (Llama-3.2-3B): Distance Histograms per Prompt Subset

Classifier Report: Binoculars

Threshold swept for fpr_0_5pct on the validation split = 0.0755; scores read from *_binoculars (lower_is_ai).

Binoculars: By Prompt Subset

Prompt Subset	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.3139	0.0003	0.0041	0.4981	0.0005
✔️ direct_reference	1,778	0.3487	0.0000	0.0045	0.4978	0.0000
indirect_reference	2,330	0.2993	0.0009	0.0043	0.4983	0.0017
❗ revise	2,016	0.2791	0.0000	0.0040	0.4980	0.0000
rewrite	1,644	0.3384	0.0000	0.0036	0.4982	0.0000

Binoculars: By Bin

Bin (cosdist)	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.3139	0.0003	0.0041	0.4981	0.0005
✔️ cosdist 0.0004953 to 0.1037	1,944	0.3502	0.0000	0.0031	0.4985	0.0000
cosdist 0.1037 to 0.1989	1,940	0.3026	0.0000	0.0052	0.4974	0.0000
cosdist 0.1989 to 0.3453	1,964	0.3062	0.0000	0.0051	0.4975	0.0000
❗ cosdist 0.3453 to 1.032	1,920	0.2939	0.0010	0.0031	0.4990	0.0021

Binoculars: Score Histograms per Prompt Subset

Binoculars: Distance Histograms per Prompt Subset

Manually Specified Full Report

This section is hardcoded: it reports the first configured classifier (EditLens Roberta-Large Score) over the generator model/sampling configurations that produced the AI side of each pair.

Generator Config	N	AUROC	TPR	FPR	Accuracy	F1
Overall	7,768	0.9219	0.4706	0.0062	0.7322	0.6374
❗ Meta-Llama-3.1-8B-Instruct-AWQ-INT4 (Temp: 0.6)	1,150	0.8603	0.3826	0.0087	0.6870	0.5500
Meta-Llama-3.1-8B-Instruct-AWQ-INT4 (Temp: 1.25)	1,130	0.9008	0.2673	0.0053	0.6310	0.4200
✔️ Ministral-3-8B-Instruct-2512-AWQ-4bit (Temp: 0.6)	1,206	0.9974	0.8027	0.0066	0.8980	0.8873
Qwen3-8B-AWQ (Temp: 0.7)	972	0.8775	0.3663	0.0062	0.6800	0.5337
gemma-4-E4B-it (Temp: 0.7)	1,010	0.9025	0.3366	0.0040	0.6663	0.5022
granite-4.1-8b-AWQ-INT4 (Temp: 0.6)	1,132	0.9393	0.5389	0.0088	0.7650	0.6963
granite-4.1-8b-AWQ-INT4 (Temp: 1.25)	1,168	0.9564	0.5479	0.0034	0.7723	0.7064