Generative artificial intelligence: Difference between revisions
removed some claims with weak/absent citations (I think for a service as large as github, a github discussion with 34 upvotes and 20 participants is not sufficient to demonstrate a substantial public backlash). the deviantart one also needs some work to this effect |
m Minor revisions for ease of reading, citation error |
||
| Line 34: | Line 34: | ||
|'''GitHub''' | |'''GitHub''' | ||
|Scraping all users' public code repositories without prior notice nor consent for a large language model ('''GitHub CoPilot'''), then locking the ability to opt-out behind a subscription service | |Scraping all users' public code repositories without prior notice nor consent for a large language model ('''GitHub CoPilot'''), then locking the ability to opt-out behind a subscription service | ||
|Labeled on the FAQ for GitHub CoPilot,<ref>{{Cite web |last=CoPilot |title=FAQ |url=https://copilot.github.trust.page/faq?s=v2qe7voltpwtv2usl4ikhs#ip-and-open-source |access-date=Mar 30, 2025 |website=github}}</ref> users who pay for either a ''Pro'' or ''Enterprise'' tier plan do not have their repositories (''repos'') scanned for the purposes of training CoPilot. There are theories that private repos may not be used for training purposes,{{Citation needed}} but these have not been verified{{CitationNeeded}}. | |Labeled on the FAQ for GitHub CoPilot,<ref>{{Cite web |last=CoPilot |title=FAQ |url=https://copilot.github.trust.page/faq?s=v2qe7voltpwtv2usl4ikhs#ip-and-open-source |access-date=Mar 30, 2025 |website=github}}</ref> users who pay for either a ''Pro'' or ''Enterprise'' tier plan do not have their repositories (''repos'') scanned for the purposes of training CoPilot. There are theories that private repos may not be used for training purposes,{{Citation needed}} but these have not been verified{{CitationNeeded}}. | ||
|[[GitHub]] | |[[GitHub]] | ||
|- | |- | ||
|'''DeviantArt''' | |'''DeviantArt''' | ||
|Training on all user-published content without prior notice nor consent for a proprietary image-generation model ('''DreamUp''')<!-- Considering the over 2 year long history that continues to have new drama stir from this, we should look into eventually making a dedicated article focused on DreamUp --> | |Training on all user-published content without prior notice nor consent for a proprietary image-generation model ('''DreamUp''')<!-- Considering the over 2 year long history that continues to have new drama stir from this, we should look into eventually making a dedicated article focused on DreamUp --> | ||
|While more speculative, it is reasonable for users to assume<ref>{{Cite web |title=What I think about DeviantArt's AI Choices |url=https://www.reddit.com/r/AI_Generator_Guide/comments/167sbit/what_i_think_about_deviantarts_ai_choices/ |url-status=live |access-date=31 Mar 2025 |website=[[Reddit]]}}</ref> that when [[DeviantArt]] initially automatically opted all users into allowing their work to be training data for generative AI,<ref>{{Cite web |date=11 Nov 2022 |title=deviantart's AI art program (and how to opt out) |url=https://www.deviantart.com/izzy-paw/journal/deviantart-s-AI-art-program-and-how-to-opt-out-936581886 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |last=Erdine |first=Önder |date=16 Nov 2022 |title=DeviantArt DreamUp is the latest contender for the AI art crown |url=https://en.digitalreport.com.tr/deviantart-dreamup-how-to-use-ai-opt-out/ |url-status=live |access-date=31 Mar 2025 |website=Digital Report}}</ref> that all content uploaded to DeviantArt was used as training data for their DreamUp tool. However, according to statements from DeviantArt CEO Moti Levy,<ref>{{Cite web |last=Robertston |first=Adi |date=15 Nov 2022 |title=How DeviantArt is navigating the AI art minefield |url=https://www.theverge.com/2022/11/15/23449036/deviantart-ai-art-dreamup-training-data-controversy |url-status=live |access-date=31 Mar 2025 |website=The Verge}}</ref> DeviantArt did not plan or intend to train their tool based on user-generated works, and that any user-generated works that were used in their model were introduced by StabilityAI. Regardless, the introduction of DreamUp to the art sharing platform has stirred controversy on the platform<ref>{{Cite web |last=Edwards |first=Benj |date=11 Nov 2022 |title=DeviantArt upsets artists with its new AI art generator, DreamUp [Updated] |url=https://arstechnica.com/information-technology/2022/11/deviantart-upsets-artists-with-its-new-ai-art-generator-dreamup/ |url-status=live |access-date=31 Mar 2025 |website=www.arstechnica.com}}</ref> and fractured it into two distinct groups;<ref>{{Cite web |last=Duchess Celestia |date=13 Nov 2022 |title=DeviantART Just Betrayed Its Whole Community. (DreamUp AI Controversy) {{!}}{{!}} SPEEDPAINT + COMMENTARY |url=https://www.youtube.com/watch?v=IGj_3OhMrAU |url-status=live |access-date=31 Mar 2025 |website=[[YouTube]]}}</ref> those for generative AI (typically those who hold newer accounts) and those against the usage of such algorithms on an art forum (typically users who have existed on the platform for far longer.) Since the introduction of DreamUp, the platform has been cluttered with AI generated images, and staff have historically, frequently, and intentionally featured multiple users who exclusively upload GenAI content<ref>{{Cite web |date=22 Jul 2024 |title=DeviantArt Seller: StygianAI |url=https://www.deviantart.com/team/art/DeviantArt-Seller-StygianAI-1077776294 |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |date=9 Oct 2024 |title=Create on DeviantArt: VeilAI |url=https://www.deviantart.com/team/art/Create-on-DeviantArt-VeilAI-1108146133 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |date=15 Jul 2024 |title=DeviantArt Seller: ExeFelix |url=https://www.deviantart.com/team/art/DeviantArt-Seller-ExeFelix-1075192370 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref> or post content that uses generative content as a base.<ref>{{Cite web |date=9 Oct 2024 |title=Create on DeviantArt: AKoukis |url=https://www.deviantart.com/team/art/Create-on-DeviantArt-AKoukis-1108151629 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref> The majority of featured creators on the site are ones who | |While more speculative, it is reasonable for users to assume<ref>{{Cite web |title=What I think about DeviantArt's AI Choices |url=https://www.reddit.com/r/AI_Generator_Guide/comments/167sbit/what_i_think_about_deviantarts_ai_choices/ |url-status=live |access-date=31 Mar 2025 |website=[[Reddit]]}}</ref> that when [[DeviantArt]] initially automatically opted all users into allowing their work to be training data for generative AI,<ref>{{Cite web |date=11 Nov 2022 |title=deviantart's AI art program (and how to opt out) |url=https://www.deviantart.com/izzy-paw/journal/deviantart-s-AI-art-program-and-how-to-opt-out-936581886 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |last=Erdine |first=Önder |date=16 Nov 2022 |title=DeviantArt DreamUp is the latest contender for the AI art crown |url=https://en.digitalreport.com.tr/deviantart-dreamup-how-to-use-ai-opt-out/ |url-status=live |access-date=31 Mar 2025 |website=Digital Report}}</ref> that all content uploaded to DeviantArt was used as training data for their DreamUp tool. However, according to statements from DeviantArt CEO Moti Levy,<ref>{{Cite web |last=Robertston |first=Adi |date=15 Nov 2022 |title=How DeviantArt is navigating the AI art minefield |url=https://www.theverge.com/2022/11/15/23449036/deviantart-ai-art-dreamup-training-data-controversy |url-status=live |access-date=31 Mar 2025 |website=The Verge}}</ref> DeviantArt did not plan or intend to train their tool based on user-generated works, and that any user-generated works that were used in their model were introduced by StabilityAI. Regardless, the introduction of DreamUp to the art sharing platform has stirred controversy on the platform<ref>{{Cite web |last=Edwards |first=Benj |date=11 Nov 2022 |title=DeviantArt upsets artists with its new AI art generator, DreamUp [Updated] |url=https://arstechnica.com/information-technology/2022/11/deviantart-upsets-artists-with-its-new-ai-art-generator-dreamup/ |url-status=live |access-date=31 Mar 2025 |website=www.arstechnica.com}}</ref> and fractured it into two distinct groups;<ref>{{Cite web |last=Duchess Celestia |date=13 Nov 2022 |title=DeviantART Just Betrayed Its Whole Community. (DreamUp AI Controversy) {{!}}{{!}} SPEEDPAINT + COMMENTARY |url=https://www.youtube.com/watch?v=IGj_3OhMrAU |url-status=live |access-date=31 Mar 2025 |website=[[YouTube]]}}</ref> those for generative AI (typically those who hold newer accounts) and those against the usage of such algorithms on an art forum (typically users who have existed on the platform for far longer.) Since the introduction of DreamUp, the platform has been cluttered with AI generated images, and staff have historically, frequently, and intentionally featured multiple users who exclusively upload GenAI content<ref>{{Cite web |date=22 Jul 2024 |title=DeviantArt Seller: StygianAI |url=https://www.deviantart.com/team/art/DeviantArt-Seller-StygianAI-1077776294 |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |date=9 Oct 2024 |title=Create on DeviantArt: VeilAI |url=https://www.deviantart.com/team/art/Create-on-DeviantArt-VeilAI-1108146133 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref><ref>{{Cite web |date=15 Jul 2024 |title=DeviantArt Seller: ExeFelix |url=https://www.deviantart.com/team/art/DeviantArt-Seller-ExeFelix-1075192370 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref> or post content that uses generative content as a base.<ref>{{Cite web |date=9 Oct 2024 |title=Create on DeviantArt: AKoukis |url=https://www.deviantart.com/team/art/Create-on-DeviantArt-AKoukis-1108151629 |url-status=live |access-date=31 Mar 2025 |website=Deviant Art}}</ref> The majority of featured creators on the site are ones who excessively or exclusively upload AI generated content.<!-- I was scrolling through their gallery and most featured artist posts were about AI creators, I stopped my search when I reached posts that released before the generative AI controversies on the platform occurred, which had a rough stopping point of around Q4 2022. | ||
https://www.deviantart.com/team/gallery --><!-- Due to my close familiarity with the situation, yes, I developed this section a lot more than initially planned. --> | https://www.deviantart.com/team/gallery --><!-- Due to my close familiarity with the situation, yes, I developed this section a lot more than initially planned. --> | ||
| | | | ||
| Line 50: | Line 50: | ||
|'''Stack Overflow''' | |'''Stack Overflow''' | ||
|Training on all active users' content without prior notice nor explicit consent for a large language model service ('''Overflow AI''') | |Training on all active users' content without prior notice nor explicit consent for a large language model service ('''Overflow AI''') | ||
|In late July of 2023, Overflow AI was made available by Stack Overflow, a Q&A forum for programmers. The information used to train the LLMs provided by the service was taken from questions and answers left by the StackOverflow community.<ref>{{Cite news |last=StackOverflow |date=Jul 27, 2023 |title=Announcing OverflowAI |url=https://stackoverflow.blog/2023/07/27/announcing-overflowai/ |access-date=Mar 30, 2025 |work=StackOverflow blog}}</ref> This action subverts one of the website's existing policies, wherein its users are disallowed from AI generating responses to answer forum posts.<ref>{{Cite web |last=StackOverflow |title=Generative AI Policy |url=https://stackoverflow.com/help/gen-ai-policy |access-date=Mar 30, 2025 |website=StackOverflow help center}}</ref> The only way users have the capability to opt-out currently | |In late July of 2023, Overflow AI was made available by Stack Overflow, a Q&A forum for programmers. The information used to train the LLMs provided by the service was taken from questions and answers left by the StackOverflow community.<ref>{{Cite news |last=StackOverflow |date=Jul 27, 2023 |title=Announcing OverflowAI |url=https://stackoverflow.blog/2023/07/27/announcing-overflowai/ |access-date=Mar 30, 2025 |work=StackOverflow blog}}</ref> This action subverts one of the website's existing policies, wherein its users are disallowed from AI generating responses to answer forum posts.<ref>{{Cite web |last=StackOverflow |title=Generative AI Policy |url=https://stackoverflow.com/help/gen-ai-policy |access-date=Mar 30, 2025 |website=StackOverflow help center}}</ref> The only way users might have the capability to opt-out currently would be to manually delete all posts and topics before deleting their account. The effectiveness of this method remains questionable, however, considering that StackOverflow's "Get Out" clause<ref>{{Cite web |last=StackOverflow |title=How do I delete all my contributions? |url=https://stackoverflow.com/help/delete-content |access-date=Mar 30, 2025 |website=StackOverflow help center}}</ref> may allow the service to retrieve content that was previously deleted. | ||
| | | | ||
|- | |- | ||
|'''Reddit''' | |'''Reddit''' | ||
|Training on all users' content without prior notice nor explicit consent for a large language model ('''Reddit Answers''') | |Training on all users' content without prior notice nor explicit consent for a large language model ('''Reddit Answers''') | ||
|In late 2024, [[Reddit]] announced the release of 'Reddit Answers,' a large language model (LLM) that was publicly stated<ref>{{Cite web |title=Reddit Answers (Currently in Beta) |url=https://support.reddithelp.com/hc/en-us/articles/32026729424916-Reddit-Answers-Currently-in-Beta |url-status=live |access-date=31 Mar 2025 |website=[[Reddit]]}}</ref> to use content created by users to train the tool, without requiring prior consent or prior public notice. Reddit's Terms of Service explicitly allows the platform to scrape user-generated content for the purpose of training "AI and machine learning models",<ref>{{Cite web |first= |date=May 29, 2025 |title=User Agreement |url=https://redditinc.com/policies/user-agreement |url-status=live |archive-url=https://web.archive.org/web/20260101023854/https://redditinc.com/policies/user-agreement |archive-date=2026-01-01 |access-date=2026-01-05 |website=[[Reddit]]}}</ref> though this is likely limited to Reddit's proprietary dataset; it has legally discouraged third-party services from scraping the platform as per the Reddit Public Content Policy.<ref>{{Cite web |date= |title=Public Content Policy |url=https://support.reddithelp.com/hc/en-us/articles/26410290525844-Public-Content-Policy |url-status=live |archive-url=https://web.archive.org/web/20251230012504/https://support.reddithelp.com/hc/en-us/articles/26410290525844-Public-Content-Policy |archive-date=30 Dec 2025 |access-date= | |In late 2024, [[Reddit]] announced the release of 'Reddit Answers,' a large language model (LLM) that was publicly stated<ref>{{Cite web |title=Reddit Answers (Currently in Beta) |url=https://support.reddithelp.com/hc/en-us/articles/32026729424916-Reddit-Answers-Currently-in-Beta |url-status=live |access-date=31 Mar 2025 |website=[[Reddit]]}}</ref> to use content created by users to train the tool, without requiring prior consent or prior public notice. Reddit's Terms of Service explicitly allows the platform to scrape user-generated content for the purpose of training "AI and machine learning models",<ref>{{Cite web |first= |date=May 29, 2025 |title=User Agreement |url=https://redditinc.com/policies/user-agreement |url-status=live |archive-url=https://web.archive.org/web/20260101023854/https://redditinc.com/policies/user-agreement |archive-date=2026-01-01 |access-date=2026-01-05 |website=[[Reddit]]}}</ref> though this is likely limited to Reddit's proprietary dataset; it has legally discouraged third-party services from scraping the platform as per the Reddit Public Content Policy.<ref>{{Cite web |date= |title=Public Content Policy |url=https://support.reddithelp.com/hc/en-us/articles/26410290525844-Public-Content-Policy |url-status=live |archive-url=https://web.archive.org/web/20251230012504/https://support.reddithelp.com/hc/en-us/articles/26410290525844-Public-Content-Policy |archive-date=30 Dec 2025 |access-date=2026-01-05 |website=[[Reddit]]}}</ref> | ||
|[[Reddit]] | |[[Reddit]] | ||
|} | |} | ||