Wikipedia:Articles for deletion/NovelAI

NovelAI

[Hide this box] New to Articles for deletion (AfD)? Read these primers!

(Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL)

WP:NCORP: Lack of notability as a company outside of primary and Japanese sources and has no significant coverage (the article uses Twitter and the company's website as sources); WP:PROMOTION because a number of other similar commercial AI subscription services exist and have been mentioned in TechCrunch/Newswire (which this article heavily relies on as their only secondary sources) and yet this is the only one that has an article and ostensibly does not do anything particularly groundbreaking or notable (c.f. Stable Diffusion or 15.ai, which are notable); lack of citations for numerous claims and significant WP:OR; potential WP:COI due to suspicious activity when AfD/proposed deletion tags were placed on the article.

It should also be noted that there is a very large amount of WP:SPA going on with the article. Users Noble93, Cebbec78 (potentially a WP:SOCK of Noble93 due to the same naming pattern) are some editors who have created a Wikipedia account just to make edits to this one article, as well as IP accounts from 93.239.148.132, 94.22.206.174, 175.159.124.34, 2a02:9b0:802d:79d6:70f8:b154:b749:d7eb whose only edits are for this one article. The whole article reeks of blatant WP:PROMOTION with not-so-hidden efforts to conceal this using multiple accounts.

Throwaway21239 (talk) 11:34, 11 October 2022 (UTC)[reply]

Automated comment: This AfD was not correctly transcluded to the log (step 3). I have transcluded it to Wikipedia:Articles for deletion/Log/2022 October 11. —^{cyberbot I}_{Talk to my owner:Online} 11:58, 11 October 2022 (UTC)[reply]
Weak keep: Likely meets WP:GNG based on third-party coverage in Japanese and Taiwanese news sources such as Yahoo! News Japan and United Daily News; WP:NONENG states that while English-language sources are preferred on the English Wikipedia, citing non-English sources is still allowed. As a native Chinese speaker and intermediate Japanese speaker, I can provide translations if requested. I'd also like to point out that the page receives just under 2,000 pageviews per day, so clearly someone is searching for it, although that isn't evidence of notability in itself. If there are WP:OR, WP:PROMOTION or WP:COI concerns, these issues can potentially be resolved by any editors who wish to clean up the article. --benlisquare_T•C•E 12:00, 11 October 2022 (UTC)[reply]
>I'd also like to point out that the page receives just under 2,000 pageviews per day, so clearly someone is searching for it, although that isn't evidence of notability in itself.

It should be noted that the NovelAI was previously included under the "See also" pages for the very popular DALL-E and Midjourney articles, which would explain the viewcounts per day. The references have since been removed from those pages, so it would be of interest to see the subsequent change in pageviews per day. Throwaway21239 (talk) 12:05, 11 October 2022 (UTC)[reply]
You removed the links on 9 October, but the page views were still high on 10 October, so there’s probably an offsite link bringing people to the article. And these people are likely already interested in NovelAI, and so will often want to make some change to the article. And most won’t have Wikipedia accounts, so we get lots of edits by IPs and new single-purpose accounts. There’s nothing suspicious about this. (And by the way, it looks a bit odd for someone with a single-purpose account to complain about other people using single-purpose accounts.) --Zundark (talk) 14:42, 11 October 2022 (UTC)[reply]
To be fair, I did suggest to the nominator to create an account, since they mentioned on the article talk page that they were unable to start the AfD nom as an IP editor; they have had edits in the past. --benlisquare_T•C•E 22:04, 11 October 2022 (UTC)[reply]
Note: This discussion has been included in the list of Software-related deletion discussions. Shellwood (talk) 12:32, 11 October 2022 (UTC)[reply]
Delete per WP:NCORP. None of the WP:SIRS criteria seem to be met, and the article reads like a press release. As someone who has kept up with the AI landscape for years, I'd never heard of NovelAI before while I hear about Stable Diffusion nonstop, and it appears that NovelAI just uses a custom implementation of SD... How exactly is this any different from any of the other countless AI companies that offer similar services? — HackerKnownAs (talk) 13:17, 11 October 2022 (UTC)[reply]
Note: This discussion has been included in the list of Delaware-related deletion discussions. dudhhr^talk _contribs (he/they) 14:46, 11 October 2022 (UTC)[reply]
I have to add that NovelAI did exist long before Stable Diffusion, it was a mainly text gpt type service. AI landscape- you mean like image side and not language model? NAI was created as a directly respond to downfall-of-AI-dungeon. And also were involved in training a GPT-NeoX-20B, the biggest public-access language model, on the paper, not specifically the novelai itself, but its creator: Eren Doğa. Not to mention that anlatan(novelai) made a joint venture with CoreWeave, I would say coreweave are pretty important player in modern ai scene, after all, they make the public 20 billions model possible.

Anyway tho, it's true that I'm single purpose account. But what can I do? Most of us never touch wiki, it's only by the popularity in japan that someone decided to make a page for it. Depsite already existed for 1 and half years.

My counter point would be, if AI Dungeon page exist, then should the NovelAI. They are literally counterpart in the same audience. Everyone in ai-storytelling scene know about it, at least for attentive users anyway. It's usually AIdungeon, NovelAI, HoloAI, KoboldAI(open-source). This is our circle.

Also on the notable(stuff?), isn't the point of this page was that Japnness people were so mad and so atonshing that NovelAI mangaged to generate high-quailty anime-like ai image? That was the point.

I know 0 sh*t about how wiki work, so I'll leave that to the professional, and it's understable that you guys editor probably don't know much about text ai scene. But as contributor of it, this is what I had to say. And I'm pretty airhead if I were to edit this, I have no idea where to start...

Lack of citations... I gusses I kidda get that where came from. Because NAI and others-ai-text-service are pretty standalone venture, had specific audience. And nobody bother to do any publicity(service speak for itself) nor did the journalist wants to write on it or anything(not the most excited thing in the world.) Andddd public large language model aren't actually 'public'. They require $25,000 gpus to run, so only corporate can 'actually' run it. Which is why it's not boom like StableDifusion did. Thank to low-memory-architecture on difusion I suppose. Or maybe people just can't read, and were more excited on good picture. Pumegit (talk) 20:41, 11 October 2022 (UTC)[reply]
oops wrong reply location, ehh Pumegit (talk) 20:48, 11 October 2022 (UTC)[reply]
>My counter point would be, if AI Dungeon page exist, then should the NovelAI.

See WP:WAX. The existence of an article of a similar topic does not make for an argument for keeping this article. It should also be noted (per WP:MEAT) that this user has made no edits outside of the discussion. Throwaway21239 (talk) 15:37, 15 October 2022 (UTC)[reply]

The difference between NovelAI and AI Dungeon is that the latter has an abundance of significant coverage (just check the references there) while the former struggles to try to meet the standards of notability. For example, the entire "Incidents" section relies on a single reference to a Twitter post made by NovelAI's account, which makes it an unreliable primary source (WP:RSPRIMARY). At best, there could be a mention of NovelAI in the article for Stable Diffusion, but as is, it clearly is not notable enough for its own article at this time. —HackerKnownAs (talk) 22:34, 11 October 2022 (UTC)[reply]
Oh well, after talking to the developers and researchers. They don't like the wiki page as it is right now. It's quite inaccurate and don't really represent what NovelAI is. Image generation is just a side project and aren't suppose to take the main stage, because they are mainly text-generation/storytelling service. They don't mind having it remove.

I'll be taking a neutral stand now, it needs a ton of efforts(articles?) to make it actually represent NovelAI. In the mean time, the event of AI art debate in japan is quite large too, which is why this page was created to begin with. But that's not what NovelAI is. Pumegit (talk) 19:38, 12 October 2022 (UTC)[reply]
The developers aren't supposed to "like" the Wikipedia article. They don't own the Wikipedia article, nor are they allowed to exert influence over the article content. If there are information gaps, it's up to third-party sources to pick up the slack. Have them go talk to tech journalists or something, complaining about the Wikipedia article being inaccurate when the sourcing simply isn't available is tantamount to just pounding sand. The image generation takes centre stage in the Wikipedia article because that's all the third-party sourcing ever talks about; even if the text generation models should be covered more, that simply cannot happen due to the reality that sourcing is utterly lacking in that area. --benlisquare_T•C•E 20:24, 12 October 2022 (UTC)[reply]
Ah, apology. I'm not familiar with how Wikipedia works. So they require third-party sources not the first-party themselves, I see. Pumegit (talk) 06:06, 13 October 2022 (UTC)[reply]
Comment, weak delete or delete and rewrite You say (I added some emphasis for certain parts of your rationale):
1. Lack of notability as a company outside of primary and Japanese sources and has no significant coverage (the article uses Twitter and the company's website as sources) I agree about the issue with primary sources but is there any policy against non-English sources?
2. WP:PROMOTION because a number of other similar commercial AI subscription services exist and have been mentioned in TechCrunch/Newswire (which this article heavily relies on as their only secondary sources) Flawed argument, but it is almost certainly promotion because of the way it is written.
3. and yet this is the only one that has an article and ostensibly does not do anything particularly groundbreaking or notable (c.f. Stable Diffusion or 15.ai, which are notable) I don't have a reliable source but going by my Google search results, NovelAI is pretty much the first company with a successful image generation model specifically for anime.
4. lack of citations for numerous claims and significant WP:OR; potential WP:COI due to suspicious activity when AfD/proposed deletion tags were placed on the article. Valid point.
FWIW, the IP editors you have named are operating from Germany, Finland, Hong Kong, and Saudi Arabia respectively. But given their editing history, they may be either different employees sockpuppeting for the company. Tube·of·Light 05:04, 12 October 2022 (UTC)[reply]
Regarding the non-English sources, for the purpose of aiding with WP:V, I've translated a few citations into English. I have one thing I'd like to point out: One of the quotes use the word "おかしい", which if we're talking about literal dictionary definitions, would mean "strange", "suspicious", "weird"; however in the context of internet colloquial speech, it's actual real-world usage is more akin to "that's fucked up, bro", or "that's screwed up", "that's messed up". I've used a more closer-to-real-world translation for this sentence, but feel free to adjust it if anyone thinks it doesn't match the tone of an encyclopedia article. --benlisquare_T•C•E 07:58, 12 October 2022 (UTC)[reply]
Keep based on the current version. I think the coverage shown in the article more than meets GNG. Regarding Japanese sources, GNG can be met with articles in any language. The "significant coverage" does not have to be in English. ···日本穣 · 投稿 · Talk to Nihonjoe · Join WP Japan! 18:51, 12 October 2022 (UTC)[reply]
Keep, there are some reliable Chinese news sources in the entry. --dqwyy (talk) 13:50, 14 October 2022 (UTC)[reply]

Delete per WP:ORG. Quoted from WP:MULTSOURCES:

"Therefore, for example, a Bangladeshi women's rights organization from the 1960s might establish notability with just one or two quality sources, while the same is not true for a tech start-up in a major U.S. metropolitan area."

Despite being a U.S. tech start-up from a metropolitan area, nearly all of the sources listed are from non-English sources, which indicates that the company is not notable in the English-speaking business world. Non-English sources may be sufficient to establish WP:GNG as pointed out above, but quoted from WP:ORGCRIT:

"These criteria, generally, follow the general notability guideline with a stronger emphasis on quality of the sources to prevent gaming of the rules by marketing and public relations professionals. The guideline, among other things, is meant to address some of the common issues with abusing Wikipedia for advertising and promotion. As such, the guideline establishes generally higher requirements for sources that are used to establish notability than for sources that are allowed as acceptable references within an article."

Right now, the Wikipedia article is written like a promotional piece: an excessive number of weak sources to try to substantiate the Wikipedia article per WP:TOOMANYREFS while not saying much about the company at all. It evidently does not meet notability standards for companies and organizations as outlined in WP:ORGCRIT. Moreover, quoted from WP:SIRS:

"Quantity does not determine significance. It is the quality of the content that governs. A collection of multiple trivial sources do not become significant. Views, hits, likes, shares, etc. have no bearing on establishing whether the coverage is significant. Similarly, arbitrary statistics and numbers (such as number of employees, amount of revenue or raised capital, age of the company, etc.) do not make the coverage significant. For the coverage to be significant, the sources must describe and discuss in some depth the treatment of the employees or major changes in leadership instead of just listing the fact that the corporation employs 500 people or mentioning that John Smith was appointed as the new CEO."

Several sources listed in the Wikipedia article seem to discuss NovelAI, but reading through the actual sources themselves, many of them discuss AI-generated anime art in general, not the company specifically, and only offer fleeting mentions of NovelAI for only a few sentences. A single-sentence mention in an article about a related topic or company is not significant coverage. When you take all of this into consideration and the suspicious circumstances surrounding the Wikipedia article pointed out above, it does look likely that this is simply another case of an AI/crypto company trying to promote their business by putting themselves on Wikipedia.

I'd like to comment that maybe there should be a separate stricter notability standard for AI companies than WP:NCORP, like WP:NCRYPTO which is used for cryptocurrency projects. There has been a lot of AI companies that have tried to put themselves on Wikipedia recently that rely on numerous sketchy sources and have clear COI violations, and so it might not be a bad idea to have something similar to NCRYPTO for AI, seeing as how the two seem to overlap frequently. UnstableDiffusion (talk) 21:28, 14 October 2022 (UTC)[reply]

Keep: Sources such as Kai-you, Sohu, ITmedia News, Gigazine, United Daily News, 4Gamers, Gamespark, FayerWayer, and Real Sound describe and discuss in depth NovelAI service or company, in terms of quality as well as quantity. Some non-English sources are cited in translation, which may improve WP:VERIFY, but in principle even translation may not be necessary.--Kainioaefa (talk) 23:32, 14 October 2022 (UTC)[reply]
Keep: Maybe the company behind NovelAI is a U.S.-based company, but considering its image generator mainly targets Anime/Manga-styled artworks, it is not that strange to have more media coverage from East Asian countries, where the preference for such illustrations much higher compared to the West. From the technical PoV, perhaps NovelAI's image generator is not that remarkable something (as they just took a F/OSS AI model Stable Diffusion and trained it with images taken from Danbooru), but the quality of anime style images it can generate is quite impressive (as User:benlisquare mentioned earlier) and that's why it got such sudden spotlights from beforementioned regions. I can't agree that it doesn't meet WP:GNG just because the sources are not in English. --ParanScreen (talk) 05:42, 16 October 2022 (UTC)[reply]

Comment: I found out why this discussion seems to be full of WP:MEAT. A direct link to this AfD discussion was posted in NovelAI's public Discord server twice: once in the #no-mic and once in the #novelai-discussion channels. The link has been sitting there as the latest post in the #no-mic channel since October 12 (direct link to the posts in question post 2). This is a pretty obvious WP:COI and explains all the WP:PROMO and WP:SPA editing present here and in the article.

Screenshot 1

Screenshot 2

Screenshot 3: one of the developers of NovelAI commenting that this AfD is likely to be rejected, two days after the link was posted in the server

Throwaway21239 (talk) 09:42, 16 October 2022 (UTC)[reply]

I voted for Keep, but I didn't do it at someone's request or saw this article on a forum somewhere. This article continues to record a high number of views exceeding 1,000 times since the day after it was created.(pageviews) You yourself also have no edit history other than this article, so I don't think it's hard to imagine that.--Kainioaefa (talk) 10:27, 16 October 2022 (UTC)[reply]

I was advised to create a throwaway account specifically to create this nomination, as I prefer editing anonymously over using an account. Besides, my edit history has nothing to do with the suspicious activity of all the once-inactive accounts mysteriously making their first edits in years only after the link was posted in the Discord server. Even without the screenshots, WP:MEAT should be suspected.

If you notice from the graph you linked, the number of hits sharply decreased when the links were removed from the See also sections of the popular articles mentioned above, and continues to decrease every day. The number was cited as high as 2,000 times per day back then, but now that it's been cut in half in just a few days, I don't know what point you're trying to make. Throwaway21239 (talk) 10:42, 16 October 2022 (UTC)[reply]

Well, that's me. As of right now, I'm not defending the article, nor do I explicitly want it delete. I already express my opinion above. Most developers agree with the deletion. Your screenshot is misleading. And WP:MEAT is not true, there is no one from the NovelAI group here beside me. And I barely touch anything besides correcting the launch date.

Screenshot 1

Screenshot 2 Pumegit (talk) 10:10, 18 October 2022 (UTC)[reply]

Comment - Not voting one way or another, but if there is consensus to delete, I think draftifying would be more appropriate as the majority of the article is about NovelAI's image generation, which was only released on October 3rd, so this could be a WP:TOOSOON situation, and if the subject isn't notable now, it may be after more coverage. Waxworker (talk) 18:39, 16 October 2022 (UTC)[reply]