AI Content Help- Ask questions, discuss, comment, complain

The current policy is where it needs to be

40.51% 64 votes

The current policy did not go far enough

29.11% 46 votes

The current policy has gone too far

25.32% 40 votes

Something else? (Mention it below)

5.06% 8 votes

Poll ended 11 months ago with 158 votes.

Rainbow Dash

Maybe a little bit wolfy

The AI Policy Thread is to briefly discuss the policy, suggest amendments, and small clarifications here and there. Discussions, in depth questions, commentary, or just complaints can go here instead. Please keep the other thread on topic and dump everything else here.

Also a small unofficial poll here.

As I said before, we feel the policy is in a good spot where it is right now, and we aren’t going to uproot everything from a poll alone.

We will however take in feedback and how we can adjust the policy to bring the most good, while prioritizing the spotlight for our human artists.

That’s all for now.

To clarify on the poll:

Voting: it’s gone too far, means that the policy is too harsh on AI content.

Voting: it hasn’t gone far enough, means it is too lenient on AI content.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

Thank you for this thread :)

I’m not the one to be highly into ai related artworks, but some are really nice. Just have to be careful with it, as with everything :)

Great being the first thread reply too :3 hope your day is going well everyone :D

And I’m wolfy too 🐺🥰

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

🐴

IRL 🎠 stallion

“Didn’t go far enough” and “goes too far” are both ambiguous in terms of which direction the going is in.

Posted 11 months ago Report

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

@PUBLIQclopAccountant
Gives more leeway. :)

Posted 11 months ago Report

Link Quote Reply

The Frowning Pony

Administrator

13 blinks per minute

@PUBLIQclopAccountant
Vote “Has gone too far” to say you’d want less restrictions on AI stuff
Vote “Not far enough” to say you’d want harsher restrictions on AI stuff

As the OP infers, this is polling for temperature, not guaranteeing any sort of policy change. Public online polls are flawed for a myriad of reasons, and this site itself attempted polling for alternative policies in the past, to very not great results.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

Rainbow Dash

Maybe a little bit wolfy

Original post has been clarified. Sadly, I can’t edit the poll names themselves.

Posted 11 months ago Report

Link Quote Reply

Diego96

I do not agree with the trending page change since the upload limit and blacklisting from the Default filter are more than enough for AI generated pictures. It’s also a bit redundant given the amount of downvoting accounts dragging images further down trending but hey, who uses filters anyways

I’d love to see the ensuing chaos with new limit, though. At least for a while.

Posted 11 months ago Report

Link Quote Reply

R-tex1241

Hope for the best

Yay, we can vote!
Thank you!

Posted 11 months ago Report

Link Quote Reply

Modera

I do not agree with the trending page change since the upload limit and blacklisting from the Default filter are more than enough for AI generated pictures.

I personally never seen any AI-generated image in trending section on front page (I don’t use any custom filters, only standard ‘Default’ and ‘Everything’) so I don’t think this change actually changes anything. Unless for people who created custom filters that hides majority of high score non-ai images.

Posted 11 months ago Report

Link Quote Reply

Brokedownandmadeone

BigThirsty ponybooru.org

I’d say at the very least the limit goes a bit far on its face. At least not without some background data.
-Are there really any current issues with AI uploads?
-What’s the volume of uploads on average per day post-event?
-If we don’t include the whole week-long event where uploading was disproportionally encouraged, what is the average volume of uploads per day? (if that’s relatively easily calculable)
-If there are any deletions, how many per day? (Decimal point is acceptable if it’s that little)

Basically, if it’s maybe 1 upload a day deleted and a small fraction vs the other “regular” uploads, I don’t see there being much of an issue with it in the first place outside of odd complaints about the technology in general (and the lack of self-filtering like with too many other tags).

Posted 11 months ago Report

Link Quote Reply

Ciaran

Senior Moderator

友情は魔法だ

@Brokedownandmadeone

-Are there really any current issues with AI uploads?

Yes

-What’s the volume of uploads on average per day post-event?

As TSP noted, it can be bad. Ignoring The Event we have been, and still are seeing, seeing uploads from some users of a dozen or more ai generated images in a few minutes. The most I’ve seen is 2 a minute for several minutes, and often all but one or two of those images is deleted. In that specific case, even the uploader had no idea why they were uploading them here, but we still had to spend over an hour talking with them about their appeal.

I hope you hear me there - there are a lot of instances where the uploaders appeal these deletions and acknowledge themselves that the images did not resemble MLP characters. They even tag the images as aberrations or with tags like wat. But, “it was the best they could get”. And they wanted to share their efforts even though they themselves didn’t see any value in the image - other than that they’d spent an hour working on it, and that was the best they could get.

Normally, if someone spends an hour with a piece of paper and a pencil, the ‘worse they could get’ is at least a uniquely copyrightable work of art that resembles an MLP character.

In these cases, not even close.

Just speaking for myself, I’ve been spending so much time with these kinds of images, and these kinds of appeals, that I’m not able to get to a lot of the stuff that I really should doing here.

So…if you’ve had a report open for 2 weeks, or 2 months, and just I haven’t gotten to it yet, that’s why. Sorry. But I am confident that as soon as this policy goes into effect that will change. Dramatically.

-If we don’t include the whole week-long event where uploading was disproportionally encouraged, what is the average volume of uploads per day? (if that’s relatively easily calculable)

Just for ai generated, and including deleted images;

Month	2022 Total/Average per day	2023 Total
January	662/21	8
February	554/19	23
March	943/30	21
April (including ‘the event’	2134/71	12
May (to date)	159/31	1

April this year was, of course, exceptional, but the average number of ai generated images per day has been arithmetically increasing.

Last year this time, we had been taking an average of 2 hours to resolve each report. Currently we are taking 4. Personally, I blame all the time we are spending looking at ai generated images on a case-by-case basis. And we haven’t even gotten through the reviews/votes on all the ai generated images from March yet.

Given that a lot of the images we delete are tagged or described by the uploaders themselves as being wat or worse, I do believe that asking the uploaders curate their own uploads, and limiting those to 2 a day, will be a great help.

-If there are any deletions, how many per day? (Decimal point is acceptable if it’s that little)

I don’t have time to do another table like that - it took almost an hour - but:

From January 1 to today, 379 ai generated images have been deleted. An average of 3 per day.
Last year, from January 1 to December 31, 82 ai generated images were deleted. An average of 0.2 images per day.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

Brokedownandmadeone

BigThirsty ponybooru.org

@Ciaran
Huh. If it’s only “the smallest slight bit maybe MLP related” being uploaded, I’m understanding the issue more. In that case, maybe there can be loosened exemptions for specific uploaders similar to how general users got to edit tags? Like if the consistently upload quality works, they can apply for a “stealth” uploader status where they get an increased limit due to their quality uploads and others may get discouraged with a 1 per day if they continuously upload rule-breaking pictures. However, none could “apply” to it. Majority I could guess would sit at the default state. That way it could satisfy more of the AI uploading issues.
However, I realize this could end up with favoritism problems and still leaving a new potential workload issue. Pretty much spitballing here.

Posted 11 months ago Report

Link Quote Reply

sammykun

Bacon enthusiast

I do think the sudden change of tab from machine learning and stable diffusion to “ai content” is too vague. the correct term is machine learning generated image after all. Just my take.

Posted 11 months ago Report

Link Quote Reply

tailsopony

tailsopony

I voted something else. Here are my thoughts:

I 100% agree with auto hiding AI posts, maybe with a caveat of “if they get a score of more than 50, they get un-hidden” or some other threshold.
I agree rate limiting is a good idea, but I have worries about how it will affect the site as tech gets better (whis is happening quick! MidJourney just did a huge update! Check it out!). Basically, as Machine Learning (ML) images get better, they will become truly indistinguishable from artist posts. Some people will take advantage of this, and post images that are actually ML generated but label them as not. This is bad for two major reasons. It actually, no kidding disenfranchises real artists, and it’s bad for the future of ML. Data from this site will be used for training by somebody (either with permission or without) and if ML algorithms get fed ML generated data, you end up with degradation of models and specific kinds of over-fitting. It’s bad for everybody.

Why do I think this will happen? It’s happening on FIMFic. We have people trying to post ML content and mis-labelling it as though they are the author all the time. The only reason we know is that they brag/ask about it on the discord. We have no other way to tell. FIMFic straight up bans ML content, and now this is what they are dealing with. I don’t want this site to have that same fight–it’s not good for anybody. A better alternative is to have a clear category for ML content where people can post, and don’t feel restricted. It is probably a good idea to moderate it as you’ve said is going to happen.

Machine Learning can be shortened to ML. Please do not use AI. AI is Artificial Intelligence, and describes many, many things. It can be programmatic, machine learning, or neither. It’s like labeling all media on this site “digital media” because they are stored online. Machine learning is the correct term, and it’s both broad and specific enough to cover the needs of the site.
I like the specific tags, but I’d replace AI with ML. Your TV has AI. Super Mario world has AI. Arguably any GUI has AI. Any photoshop filter is a form of AI. This is not what we are targeting. We are targeting ML generated and assisted images.

Anyways, these are my thoughts on it.

Posted 11 months ago Report

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

@Brokedownandmadeone

-Are there really any current issues with AI uploads?

Yes

-What’s the volume of uploads on average per day post-event?

As TSP noted, it can be bad. Ignoring The Event we have been, and still are seeing, seeing uploads from some users of a dozen or more ai generated images in a few minutes. The most I’ve seen is 2 a minute for several minutes, and often all but one or two of those images is deleted. In that specific case, even the uploader had no idea why they were uploading them here, but we still had to spend over an hour talking with them about their appeal.

I hope you hear me there - there are a lot of instances where the uploaders appeal these deletions and acknowledge themselves that the images did not resemble MLP characters. They even tag the images as aberrations or with tags like wat. But, “it was the best they could get”. And they wanted to share their efforts even though they themselves didn’t see any value in the image - other than that they’d spent an hour working on it, and that was the best they could get.

Normally, if someone spends an hour with a piece of paper and a pencil, the ‘worse they could get’ is at least a uniquely copyrightable work of art that resembles an MLP character.

In these cases, not even close.

Just speaking for myself, I’ve been spending so much time with these kinds of images, and these kinds of appeals, that I’m not able to get to a lot of the stuff that I really should doing here.

So…if you’ve had a report open for 2 weeks, or 2 months, and just I haven’t gotten to it yet, that’s why. Sorry. But I am confident that as soon as this policy goes into effect that will change. Dramatically.

-If we don’t include the whole week-long event where uploading was disproportionally encouraged, what is the average volume of uploads per day? (if that’s relatively easily calculable)

Just for ai generated, and including deleted images;

Month 2022 Total/Average per day 2023 Total

January 662/21 8

February 554/19 23

March 943/30 21

April (including ‘the event’ 2134/71 12

May (to date) 159/31 1

April this year was, of course, exceptional, but the average number of ai generated images per day has been arithmetically increasing.

Last year this time, we had been taking an average of 2 hours to resolve each report. Currently we are taking 4. Personally, I blame all the time we are spending looking at ai generated images on a case-by-case basis. And we haven’t even gotten through the reviews/votes on all the ai generated images from March yet.

Given that a lot of the images we delete are tagged or described by the uploaders themselves as being wat or worse, I do believe that asking the uploaders curate their own uploads, and limiting those to 2 a day, will be a great help.

-If there are any deletions, how many per day? (Decimal point is acceptable if it’s that little)

I don’t have time to do another table like that - it took almost an hour - but:

From January 1 to today, 379 ai generated images have been deleted. An average of 3 per day.
Last year, from January 1 to December 31, 82 ai generated images were deleted. An average of 0.2 images per day.

Interesting stats :o

_{^{unrelated but I need to learn how to do those tables at some point too, looks really neat 😞🙂}}

Posted 11 months ago Report

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

@Ciaran
So your last set of stats should look like this I think:

Dateline	Total Deleted	Per Day
January 1 - December 31, 2022	82	0.2
January 1 - May 6, 2023	379	3

Posted 11 months ago Report
Edited 11 months ago because: Table worked! 🥳

Link Quote Reply

sammykun

Bacon enthusiast

@tailsopony
I agree. People are discriminating the term of ai because of the ai generated images when in reality something as simple as a calculator is a form of artificial intelligence. It should be called machine learning since it is the correct term after all. I also think that putting stable diffusion as generator is a wrong imput because stable diffusion is not a generator, it’s the tecnology itself that makes this machine learning images posible.

Posted 11 months ago Report

Link Quote Reply

Ciaran

Senior Moderator

友情は魔法だ

@Brokedownandmadeone
This is just my own anecdotal opinion, but I think most of the people uploading the very best ai generated images are already only uploading one or two a day. Even in the case where they’re uploading image sets, they seem to be spacing them out to just one or two a day already.

Maybe that’s because they want to give each image get its own prime time and thus more votes, or maybe it’s because they already have sensed that too many of the same image all at once is often perceived and downvoted - appropriately or not - as ‘spam’.

I see the same thing with pros uploading only one or two images out of giant image packs, giving each image some time to mature and get responded to before adding another. Sometimes this even builds a lot of excitement for subsequent uploads that a single bulk upload would not have.

Again, this is anecdotal, but a lot of the people that seem to me to be responsible for the spam seem to have only just discovered ai generated image creation, and it’s their first experiences with creating anything remotely resembling what they consider good art.

So even if it’s just an anime girl in a swimming suit with an anime background, they want to share it. All of it. Every single one of the best of the 10 images they could generate before their free pass on the generator ended.

And because they have no prior experience showing art to others, getting most or all of their work downvoted or deleted can feel pretty personal.

Actual artists tend to get through that phase pretty early in their learning curves. And in my experience do not seem to be the people uploading anything like ‘spam’.

Hopefully the 2 per day limit will help the new folks slow down, and curate their own output a little better.

Posted 11 months ago Report

Link Quote Reply

Ciaran

Senior Moderator

友情は魔法だ

@☬ ᏝᏆᏁᏨᎾᏝᏁᏰᏒᎬᎳᏕᎿᎬᏒᎱᎯᏁ ☬
Yeah, that’s right. I know that a change in the average from 0.2 to 3 a day may not seem like much, but it’s not that well spread out in practice, and because a lot of the worse spam is from people who’ve never dealt with having their work critiqued or downvoted or deleted before, the appeals, in my experience, seem to take a lot longer because we spend so much time going through the same questions over and over that actual artists tend to already understand.

It’s reminiscent of the reports that climate change will ‘only’ change average global temperatures a few degrees. It doesn’t seem like much on paper, but when you’re living it, it can really be big.

For me, one of the biggest advantages of the proposal is that we are standardizing our terms, and eliminating the case-by-case review of the great bulk of the problematic ai generated ‘spam’.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

Brokedownandmadeone

BigThirsty ponybooru.org

@Ciaran
Hmmm. Interdasting.

Since it seems like it wouldn’t change the usual ML/AI users, maybe you really did put in a lot more legwork into this. As a suggestion, you should probably add this sort of context in the future to assure that not much is actually changing in the big picture. Not meant as any kind of insult, but a lot of guesses and assumptions get made when valuable context like this is missing. Hindsight is 20/20 I guess. >>2006027 (just imagine I posted the last panel here. Real slim pickins for ‘hindsight’)

@☬ ᏝᏆᏁᏨᎾᏝᏁᏰᏒᎬᎳᏕᎿᎬᏒᎱᎯᏁ ☬
You should just hit “reply” for bulky posts like that instead of quote. Less scrolling needed and doesn’t add anything to quote the whole thing.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

@Brokedownandmadeone
I usually do :) but I needed to see the table formatting :(

Good point though :3

Posted 11 months ago Report

Link Quote Reply

🂾𝕃𝕚𝕟𝕔𝕠𝕝𝕟𝔹𝕣𝕖𝕨𝕤𝕥𝕖𝕣𝔽𝕒𝕟🂽

Goodbye guys ;)

@Ciaran
I think that’s a good proposal :) and you guys are handling this really well in my opinion, great job :)

Posted 11 months ago Report

Link Quote Reply

Background Pony #F65B

Just wanna remind people how much ai/ML hurts actual artists. Heck even the writer strike is having issues with AI/ML stuff.

But I feel anyone who is super into AI/ML doesn’t actually really care about artists…like at all. Your actions speak so much louder than what you say.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

Clear Vision

Petrolhead Artist

Since it’s okay to share more general thoughts in this thread…

In that specific case, even the uploader had no idea why they were uploading them here, but we still had to spend over an hour talking with them about their appeal.

This kinda highlights the biggest problem I have with people who are supportive of AI art right now to an almost delusional degree. There seems to be no real reason for it other than “just because”, or they believe that “AI is good for the future of art” without really explaining or giving any valid reason why.

I can’t speak for all artists, every artist has their own reasons for why they make art, but one of the things that I feel that applies to many artists is that they post art because they want to share something that they believe to actually be good with the community - whether it be something that expresses their own interests, the fandom’s interests, or simply something simple and funny to make people laugh. They’re not just doing it to make the total number of images on the Internet go up, but because they believe they’ve made something with value.

Having more art of a character isn’t inherently a good thing if it’s all just low effort images with no soul or thought put into it. And unfortunately there’s people who think more is always better no matter what.

Machine Learning can be shortened to ML. Please do not use AI. AI is Artificial Intelligence, and describes many, many things. It can be programmatic, machine learning, or neither. It’s like labeling all media on this site “digital media” because they are stored online. Machine learning is the correct term, and it’s both broad and specific enough to cover the needs of the site.

Why is terminology an issue here when the tag is already aliased and most people refer to it as AI art anyway? Even the U.S Copyright Office and the UK Government? Yes the term AI is broad and yes machine learning more accurately describes what is actually happening, but it’s not incorrect to call it AI generated art as machine learning is an application of AI. Even if the alias was swapped around, it would function exactly the same.

To use your analogy, all media on Derpibooru technically is digital. Even if you look at the stuff under the traditional art tag, you’re seeing a digital image rather than the physical canvas or sheet of paper itself. It’s not unusual for editing to be necessary after scanning traditional art for the resulting digital image to more closely match the actual physical art.

@Background Pony #F65B

But I feel anyone who is super into AI/ML doesn’t actually really care about artists…like at all.

Absolutely. It doesn’t apply to everyone who has ever used AI content, but definitely to the “AI art is the future” types.

Posted 11 months ago Report
Edited 11 months ago

Link Quote Reply

tailsopony

tailsopony

Terminology is important because that’s how language works and ideas develop. If I’m running a website that categorizes tissue paper, and I decide to label all Kleenex as tissue paper, but not Puffs, then actually searching for Kleenex or puffs becomes unnecessarily complicated. They are both tissue paper. Basically, all ML generated images are AI, but not all AI images are ML. AI and ML are not the same thing, and as this societal conversation goes on it will be more and more important for the layman to understand that. In fact, literally anything on this site that was processed in photoshop, saved as a jpeg, or had some sort of automated interpolation performed should be tagged as at least “AI assisted” if we are going to use “AI assisted” as a tag.

But that’s not useful to the user, and people don’t currently have an issue with an artist using a photoshop filter to add texture, use a digital paintbrush tool, or changing the alpha levels to adjust lighting. People are concerned about the machine learning aspect, not the artificial intelligence aspect, and it’s important for the layman to understand this as ML and AI explode into everyday life in all these new ways.

in this specific use case, I would argue that it’s important because we don’t know how novel AI might actually be applied. There might be new techniques which aren’t ML that still have AI generated data/images that will need to be labelled. Somebody could write a programmatic image generator–not powered by ML at all, not trained on “real” data, and would be forced to lump it in with ML art when it might be the “vegan” option that doesn’t feed off of artists.

I don’t know how this will evolve. No one does. So it makes sense to use the most accurate term for the moment, especially when it does fully encompass the current hubub. There’s no need to classify all regular shapes as polygons when we can easily call them squares, triangles, or hexagons as needed. Especially when people are really pissed off about squares right now and we just decide that they are actually pissed off about polygons in general because we’ve never had a geometry class, but don’t worry about triangles and hexagons–they’re not classified as polygons here.

But, like, that’s just my opinion man.

Posted 11 months ago Report

Link Quote Reply

Interested in advertising on Derpibooru? Click here for information!

Pony Arts & Prints!

Help fund the $15 daily operational cost of Derpibooru - support us financially!