Please ban AI generated posts

[jcf]

A growing problem, especially in regards to Facebook, is eejits screaming "AI" when
an image or photograph is posted of something with which they are unfamiliar. I've
seen some double-down even after the photo/illustration/painting etc, is shown to be
"real" and predate AI. This often occurs with manufacturer concept art, brochure
illustrations, model and mockup photographs, photographs of lesser known aircraft and
testbeds i.e. the XF-88B supersonic propeller testbed. The most eyerolling, and rather
hilarious, declarations are those made against work by well known artists, illustrators and
designers. The worst part is that, in most cases, a simple 15 second image search will
disprove the accusation yet they clearly can't be arsed to do so, cleary preferring their
smug, kneejerk declaration to gaining even a smidgen of knowledge.

AI = Affirmation of Ignorance.

 

[lastdingo]

I think a ban on AI-generated images is fine, with exceptions for AI-themed threads and images clearly marked (not just watermarked) as AI content.

Regarding AI texts; members who happen to post a lot of very bad texts (AI or else - this saves the verification issue) could get escalating temporary posting bans. 1 day, 2 days, 3 days ...
Mods should define "bad" beyond mere "disagreement" or mistaken claims, though.

I suspect there will soon be plugins for forum operators that permit a somewhat reliable AI detection. The AI slop text problem is so great in general that it's all but guaranteed that huge effort is being put into finding a technical solution.

 

[dannydale]

A growing problem, especially in regards to Facebook, is eejits screaming "AI" when
an image or photograph is posted of something with which they are unfamiliar. I've
seen some double-down even after the photo/illustration/painting etc, is shown to be
"real" and predate AI. This often occurs with manufacturer concept art, brochure
illustrations, model and mockup photographs, photographs of lesser known aircraft and
testbeds i.e. the XF-88B supersonic propeller testbed. The most eyerolling, and rather
hilarious, declarations are those made against work by well known artists, illustrators and
designers. The worst part is that, in most cases, a simple 15 second image search will
disprove the accusation yet they clearly can't be arsed to do so, cleary preferring their
smug, kneejerk declaration to gaining even a smidgen of knowledge.

AI = Affirmation of Ignorance.

I feel that facebook and its flood of ai slop propaganda has left a great many people traumatized, and this is part of what the trauma response looks like.

 

[apparition13]

This isn't directed at AI in particular, but I pretty much avoid the general forums ("aviation and space" and "military") now because it's page after page of mostly useless speculation. A consequence of which is that the "what's new" link has also become mostly useless since it's now filled with posts from those forums rather than what I think of as the heart of the site, the secret projects section.

 

[Jemiba]

Posted by jsport, and deleted as quite obviously at least „modified“ by AI, I put it up here again for discussion, as to me it looks as if just the text was dealt with via AI … as long, as nobody proves, that one of the shown variants is just AI fake, too. And yes, a big part of the problem was the omission of the one who tried to translate it by AI to do a proper check of the results !
Incorrect operation of Autotranslation often is a problem here, resulting in posts in languages, that only a minority of the members can read.
In our example here, I think that fault OCR at least was a main issue, but IF the shown graphics/pictures are real and not hallucinated (I‘m not an expert, and cannot decide on this), it may be an interesting document anyway. I‘m honestly not sure how to deal with it.
Opinions ?

 

[overscan (PaulMM)]

Well, with the text being broken, it makes the diagram useless.

If we can find the 'good original', then by all means post that, in whatever language it is.

 

[overscan (PaulMM)]

It doesn't seem to be a translation of a good original, but an AI original.

I asked AI (Gemini) about the diagram and it said

You are completely right. This diagram is actually an AI-generated image that mimics the style of a real military vehicle infographic.
The image uses realistic layouts, colors, and vehicle silhouettes, but if you look closely, it is filled with visual distortions and nonsensical details.
Here is what gives it away as AI-generated:

• Garbled Text: While it successfully creates the appearance of acronyms and numbers (like M1126, DVH, ICV), most of the smaller labels, dates, and descriptions are illegible, warped, or completely invented gibberish.
• Mismatched Vehicle Details: The small vehicle graphics look plausible from a distance, but they lack the distinct, accurate technical details of real Stryker variants (such as specific weapon turrets or sensor packages).
• Illogical Connections: The "evolutionary tree" lines connect boxes in ways that look visually interesting but do not represent actual historical military development timelines.

If you are looking for actual, verifiable information on how the Stryker vehicle family evolved, let me know!
If you'd like, I can find or explain:

• The real historical development timeline of the Stryker.
• The actual technical differences between the original flat-bottom hull and the Double V-Hull (DVH).
• A specific list of real Stryker variants (like the M1126 ICV or the M1296 Dragoon).

Let me know what specific Stryker details or variants you are interested in exploring!

 

[helmutkohl]

Posted by jsport, and deleted as quite obviously at least „modified“ by AI, I put it up here again for discussion, as to me it looks as if just the text was dealt with via AI … as long, as nobody proves, that one of the shown variants is just AI fake, too. And yes, a big part of the problem was the omission of the one who tried to translate it by AI to do a proper check of the results !
Incorrect operation of Autotranslation often is a problem here, resulting in posts in languages, that only a minority of the members can read.
In our example here, I think that fault OCR at least was a main issue, but IF the shown graphics/pictures are real and not hallucinated (I‘m not an expert, and cannot decide on this), it may be an interesting document anyway. I‘m honestly not sure how to deal with it.
Opinions ?

its got the Gemini AI logo in the bottom right corner.
On the topic of AI, I keep deleting the posts of this new member, who keeps attempting to post AI pics of Trump as a cat and the F/A-XX

 

[Dilandu]

For me AI is mainly a good illustration tool, helping greatly to turn sketches and Paint drawnings into schemes and images. Also its good for preliminary discussions, for finding flaws in concepts. Its also good for explaining articles written on foreign language, when auto translator failed to grasp technical details sufficiently to understand. I wouldn't rely on AI to wrote a text, though; first of all, it qould be very generic and bland, second - it could easily contain mistakes.

 

[jsport]

Posted by jsport, and deleted as quite obviously at least „modified“ by AI, I put it up here again for discussion, as to me it looks as if just the text was dealt with via AI … as long, as nobody proves, that one of the shown variants is just AI fake, too. And yes, a big part of the problem was the omission of the one who tried to translate it by AI to do a proper check of the results !
Incorrect operation of Autotranslation often is a problem here, resulting in posts in languages, that only a minority of the members can read.
In our example here, I think that fault OCR at least was a main issue, but IF the shown graphics/pictures are real and not hallucinated (I‘m not an expert, and cannot decide on this), it may be an interesting document anyway. I‘m honestly not sure how to deal with it.
Opinions ?

My bad, didnt pay close attention. pictures are ungenuine, worse than useless

 

[quellish]

I suspect there will soon be plugins for forum operators that permit a somewhat reliable AI detection. The AI slop text problem is so great in general that it's all but guaranteed that huge effort is being put into finding a technical solution.

For this forum, probably not. Many members use the formal, technical prose that AI LLVMs are trained on, leading to many false positives.

There is also an issue where SPF is the primary source of information for many obscure topics these same systems have been trained on . Again, this creates false positives.

An automated “AI detector” is probably not going to work well here.

 

[quellish]

For me AI is mainly a good illustration tool, helping greatly to turn sketches and Paint drawnings into schemes and images. Also its good for preliminary discussions, for finding flaws in concepts. Its also good for explaining articles written on foreign language, when auto translator failed to grasp technical details sufficiently to understand. I wouldn't rely on AI to wrote a text, though; first of all, it qould be very generic and bland, second - it could easily contain mistakes.

Currently, there is no AI system that does the research and other work that members do, or has the experience of some members.

As recently posted in another thread questions asked to “frontier” AI LLVMs about the “SWERVE” program yielded empty, wrong, or incorrect responses when correct information is publicly available and in some cases posted here on SPF.

I have seen one LLVM regurgitate categorically incorrect information posted to SPF about a very specific but obscure topic.

 

[Yellow Palace]

As much as I despise LLMs being rammed down our throats by software companies hoping that they can find a way to make them profitable, I really don't think there's an effective way to ban them without culling a significant proportion of the forum. The overlap between the best LLM-generated text and the worst human-generated text is considerable. There's a signifcant class of the latter that I feel the forum would also be better off without, but banning low-quality posts would be almost totally unenforceable, require a huge amount of moderator time, and fundamentally change the forum's character.

There's also the consideration that this is an English-language forum. A respectable number of users here are non-English-speaking, and rely on machine translation. Which, these days, is done by LLMs, or at least LLM-like tools. All things considered, the forum is better off for their contributions.

 

[Dilandu]

There's also the consideration that this is an English-language forum. A respectable number of users here are non-English-speaking, and rely on machine translation. Which, these days, is done by LLMs, or at least LLM-like tools. All things considered, the forum is better off for their contributions.

Just an example. I'm reasonably fluent with English. I knew nothing about Swedish. When I researched data about Rb 08 missile, I was forced to use autotranslate to work with documents. But - in some cases, standard Google autotranslate could not translate technical details in such way, so I could understand them.

In such cases, the explanation is required - which LLM's like Gemini could provide. I put the fragment (on Swedish) that got me confused into Gemini, asked it to "translate and explain details", and things immediately clicked in places. Just a little explanation behind the confusingly-translated phrases, and they started to make sence.

 

[Avimimus]

There are lots of social and other reasons to... well, support the Butlerian Jihad...

But I'd suggest focussing this discussion on:
1) How does AI upscaling interfere with our ability to understand aircraft? Should upscaled images be banned?
2) Should there be a rule against people sharing information obtained through LLMs (without providing primary or secondary sources), and should that rule include the ability to ban users spreading misinformation?

 

[overscan (PaulMM)]

Just an example. I'm reasonably fluent with English. I knew nothing about Swedish. When I researched data about Rb 08 missile, I was forced to use autotranslate to work with documents. But - in some cases, standard Google autotranslate could not translate technical details in such way, so I could understand them.

I've found that Google Translate is pretty awesome with translating Russian manuals etc, and drastically better than 20 years ago, but Chinese seems harder. Surprised that Swedish was an issue.

In such cases, the explanation is required - which LLM's like Gemini could provide. I put the fragment (on Swedish) that got me confused into Gemini, asked it to "translate and explain details", and things immediately clicked in places. Just a little explanation behind the confusingly-translated phrases, and they started to make sence.

This is a perfectly fine use of AI tools. If you then make a post on the forum using the AI enhanced translation, then fair play to you.

If however you just ask Gemini "tell me about the Rb08 missile" then copy paste the answer here, then that is the definition of null content, like copy pasting the wikipedia article.

This is happening a fair amount. I think some of it is probably bots, using AI to try to create posts that might vaguely seem like genuine posts and bypass the approval step, but some does just seem to be actual people.

 

[Dilandu]

I've found that Google Translate is pretty awesome with translating Russian manuals etc, and drastically better than 20 years ago, but Chinese seems harder. Surprised that Swedish was an issue.

Its extremely good for most cases, but in technical details it sometimes could be confusing.

 

[Damirtsar]

Or ai generative images

 

[Scott Kenny]

Or ai generative images

If they're labeled as AI that's one thing. I greatly struggle to get ideas in my head out to show other people.

But anyone knowingly posting AI images without labeling them needs a warning and probably a vacation from the forum. (Yes, I'm still mad about being tricked by some pre-AI publicity photos that the XFV-12 actually flew.)

 

[Dilandu]

But anyone knowingly posting AI images without labeling them needs a warning and probably a vacation from the forum. (Yes, I'm still mad about being tricked by some pre-AI publicity photos that the XFV-12 actually flew.)

Agreed, because AI capability of realistic images is... hard to underestimate, especially if images are done in retro style (black & white, intentionally low quality). As author of several highly realistic technical alternate history works - which were mistook for real things despite being clearly stated to be alternate history! - I realize quite well the confusion, which carefully-made AI image could cause in technical areas. So I'm fully agree with the need to label AI images on Secretprojects.

If they're labeled as AI that's one thing. I greatly struggle to get ideas in my head out to show other people.

Same for me; my artistic skills are pretty moderate, and AI provide a perfect way to visualize what I'm trying to say.