-

2024-09-27 08:13:03

Ciara

CiaraNi@mastodon.green

Another thing that AI is ruining: a trusting nature. I’ve caught myself hesitating before I boost a superb photo that’s not obviously AI-generated but may be too good to be true. I check the AltText for signs of individuality like ‘I’ or ‘our local park’ or a random aside. I like fingertips at photo edges. I like typos and garbled grammar and messy phrasing. These are now my Captchas, a human fingerprint pressed into words or photos like a wax seal that verifies: this is from a real person.

#AI

show all 31 comments

2024-09-27 08:18:17

John Arnold

JohnDA@social.vivaldi.net

@CiaraNi Alas, I‘ve heard that some people include intentional misspellings in AI-generated text just in case readers delvve into analysing it along the lines you mention. I shouldn’t wonder if added ‘features’ will be added to images as well.

2024-09-27 08:23:37

Monsieur Ralph

RalphBassfeld@swiss.social

@CiaraNi I’m sure there is a prompt for that. 😬

2024-09-27 08:33:26

Annie

theappletree@ruhr.social

@CiaraNi That's very sad to hear but absolutely understandable. I've also hesitated once or twice already.

2024-09-27 08:34:10

Sebbe

royalrex@mastodon.online

@CiaraNi if you haven't, read Filterworld by Kyle Chayka. He has en example of how recommender algos is shaping the mainstream on SoMe, and that turns into the example of people adopting jokes as if it was their own kids or friends doing it. Simple way to game SoMe to get likes etc. Seems like the same concept as faking the human touch. It's... Depressing...

2024-09-27 09:02:44

Lykso

lykso@tiny.tilde.website

@CiaraNi Which will work until that also gets mimicked. I think we're going to have to fall back on personal relationships and reputation. I dunno that this Web of strangers can survive.

2024-09-27 09:04:08

🍵 holly 🇿🇦⛵️🇸🇽

holly@social.v.st

@CiaraNi @juliette Yep, my partner took an astounding photo of Mt. Fuji out the window of the shinkansen and some guy kept accusing it of being fake. :/

2024-09-27 09:05:44

Stuart Gray

StuartGray@mastodonapp.uk

@CiaraNi you should be wary of adopting such a simplistic binary approach.

Using Generative AI to produce alt text has a huge take up & use among the disabled, blind, and visually impaired community.

If you browse the right hashtags you’ll find it’s a huge transformative benefit for them.

The tools are neutral, it’s the (bad faith) actors that should be verified & if necessary blocked.

2024-09-27 10:04:46

Mans R

mansr@society.oftrolls.com

@CiaraNi Spectacular scenes created entirely in Photoshop have existed for decades, sometimes even winning prizes.

2024-09-27 10:59:49

Trammell Hudson

th@social.v.st

SECRET document (a list of branch 31 military personnel in 1945) with part of a hand holding it flat for photographing

@CiaraNi fingertips on the edges are also how you identify quality FOIA results

2024-09-27 13:29:53

Electropict

electropict@mastodon.scot

@CiaraNi

Mmmmmmmm. But most of the weird phrasings and a substantial proportion of typos I’ve seen in recent years are the result of humans who do know better but don’t notice what their autocorrect is up to.

2024-09-27 16:55:00

Petra van Cronenburg

NatureMC@mastodon.online

@CiaraNi As a journalist, I don't trust photos since Photoshop. 😉
But it gets harder and harder to recognise professional fakes. And in social media, I only boost photos if they were not stolen from photographers. Unfortunately there are accounts collecting followers with the work of others, pretending that the photos were their own works. This existed before AI. Now AI floods everything.

2024-09-27 17:00:32

Arratoon

arratoon@beige.party

@CiaraNi Unless I can see all six fingers I don’t trust it…

2024-09-27 18:54:38

reynir

reynir@social.data.coop

@CiaraNi I made a call to some company or other organization. The voice at the other end was too "perfect" I thought so I struggled figuring out if I was talking to an actual person or not. I was convinced when she said something random to me hesitating so much when speaking (I didnt want to have a conversation with a bot…)

2024-09-27 19:35:39

Lizzie she/her - also we/us

Lizette603_23@mastodon.social

@CiaraNi " I like fingertips at photo edges. I like typos and garbled grammar and messy phrasing" you're gonna LOVE me

2024-10-27 19:48:31

Ciara

CiaraNi@mastodon.green

Is there a polite way to ask if an amazing photo is real or AI? I just refrained from boosting a fab photo because it seems borderline. I chose caution because the AltText is generic with no placename. But I'm not certain. It may be real.

I hate the way #AI has enshittified this too, making me suspicious of simple things. Making me perhaps unfairly think someone posts fake images when they may just be a good photographer. Making me feel I must do homework and research before sharing a photo.

2024-10-27 19:52:23

Boab

boab@mastodon.gamedev.place

@CiaraNi funny you asked, fist I saw your fantastic Alt texts I thought they were AI generated, because of the level of details.

2024-10-27 19:52:54

Boab

boab@mastodon.gamedev.place

@CiaraNi btw let me know if that was unpolite ;)

2024-10-27 19:55:56

kel

kel@mastodon.online

@CiaraNi

Yeah, same here!

I tend not to any more, it's sad, but just how life is at the moment.

I think all such materials should carry an Identifier, somehow, NO-AI or AI as a tag would be sufficient.

That's a social thing though, based on trust and uptake.

Could be on an piece by piece basis or in profile info.

I am going to add NO-AI to mine.

Although... it is increasingly difficult to avoid, or even *know* ...

2024-10-27 19:58:15

Ciara

CiaraNi@mastodon.green

@boab That's interesting! I write too-long #AltText because I write them as if I were describing the scene off-the-cuff to a blind friend standing beside me, so I ramble and make asides, like I do in conversation. It never struck me that lots of detail could seem AI-generated, but I can see how.

Funnily enough, I suspect AI in the opposite case, when AltText has little detail with no placename or personal touch, just 'sunrise at frozen lake'. Bloody AI makes us suspicious in every direction!

2024-10-27 20:03:14

katzenberger 🇺🇦

katzenberger@mastodon.de

@CiaraNi

"Where was this picture taken?"

Assume the best, while allowing them to disappoint you.

2024-10-27 20:19:37

Lizzie she/her - also we/us

Lizette603_23@mastodon.social

@CiaraNi Or you could share with the rider "Don't know if it's tampered with, but it made me happy so I'm sharing" That's what I've done. No one has put a hit out on me yet for doing so.
And I, too, hate this enshittification.

2024-10-27 20:27:44

Vincent

vincent354@mastodon.green

@CiaraNi I don't think there is. I tend to say something like "Thanks Midjourney!" and see what comes back. If not sure... I don't engage.

But yeah. It's not possible to trust anything...

2024-10-27 20:32:17

Stuart Gray

StuartGray@mastodonapp.uk

@CiaraNi Not sure but the simplest approach could be to compliment the photographer and ask what their workflow was?

Bear in mind it’s *very* rare to see any photo that hasn’t undergone some degree of post processing, even if that’s just sharpening or colour correction, and like any art form more “artistic” images can involve more processing or refinement - few pro photographers publish raw photos only.

2024-10-27 20:47:28

Allen

allenstenhaus@beige.party

@CiaraNi My apologies if this comes across as argumentative in any way, but please don't take typos and grammar errors as proof of humanity. Scammers program their bots to intentionally make common spelling errors specifically to make them seem more human to trick people into being scammed.

That aside, I do fully agree with the point you are making. It's so hard to know what is real and what isn't. It's to the point now where I'm beginning to assume everything is fake. My former healthy skepticism is morphing into a blend of nihilism and distrust, neither of which I like.

Of course, we could argue the internet has always been fake to some degree, but now? I spend less and less time online than I used to specifically because nearly everything online is fraudulent in some way.

This is not healthy.

2024-10-27 20:49:40

Nonya Bidniss :CIAverified:

Nonya_Bidniss@infosec.exchange

@CiaraNi There's been many a time that I'll take that too-good-to-be-true pic and do a TinEye or Google Lens search to make sure. Almost always it turns out to be uncredited (scraped) and often AI. I've got kind of a spidey sense about it now.

2024-10-27 21:32:14

Mike Nelson Pedde

wolfnowl@universeodon.com

@CiaraNi

It can be very hard to tell, and I'm not against AI art per se (stealing people's work to train LLMs is a separate issue). For myself, I add a #NotAI hashtag to my images. 🤷‍♂️

But if you want to know, ask!

2024-10-27 22:11:45

Jupiter Rowland

jupiter_rowland@hub.netzgemeinde.eu

@Ciara @Boab I guess there are enough signs that my image descriptions are hand-written, especially for my original virtual world renderings.

Alt-texts which lately keep reaching exactly 1,500 characters or only few characters short of that limit.
Alt-texts that also mention an even longer image description in the post. And there is an even longer image description in the post. Who asks an AI to describe an image in lots of details and then again in even more details?
No AI can produce image descriptions with five-digit character counts like the long one in the post.
Excessive detail information about an absolutely obscure niche topic in the long description.
Description of visual details that aren't visible at the image's resolution.
Transcripts of text that isn't legible or not even visible at the image's resolution.
Sometimes I run an extra thread with an image-describing log.

#Long #LongPost #CWLong #CWLongPost #AltText #AltTextMeta #CWAltTextMeta #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

2024-10-27 22:15:37

Jupiter Rowland

jupiter_rowland@hub.netzgemeinde.eu

@katzenberger @Ciara Ask me, and I'll tell you in over 1,000 characters because I have to.

Show me one AI that does that.

#ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

2024-10-28 06:09:34

Ciara

CiaraNi@mastodon.green

@jupiter_rowland That's great commitment to AltTexting!

2024-10-28 12:07:33

Jupiter Rowland

jupiter_rowland@hub.netzgemeinde.eu

@Ciara And yet, there are people trying to talk me out of it. For example, I shouldn't transcribe text that's so tiny that it isn't even recognisable in the image as text because it's only a blob of a dozen pixels. They say that 40,000 or 60,000 characters of description and explanations for one image are too much.

The only thing I'm reconsidering myself currently is whether to keep these monster descriptions in the post or put them into external documents and link to them.

#Long #LongPost #CWLong #CWLongPost #ImageDescription #ImageDescriptions #ImageDescriptionMeta #CWImageDescriptionMeta

2024-10-29 18:56:03

Ciara

CiaraNi@mastodon.green

@jupiter_rowland 60,000 characters is certainly commitment!