AI fails

[briantrumpet]

I'm sceptical. So I need somewhere to put AI fails I've spotted.

First up, I saw a FB post publicising a concert in Exeter Cathedral, and the 'photo' didn't look quite right. No idea why they didn't just use a proper photo, as it's not hard to get a good one. (The real photo is mine.)

 

[HMS_Dave]

I've dabbled with this stuff. I use Krita with the AI diffusion plug-in so my Graphics card does all the computation on my local device.

The results speak for themselves.

This one is modeled on a few key words i would describe as a brexiteer, cyclist, Heavy Keyboard user, trolling on a popular cycling forum...

The results speak for themselves.

 

[Pblakeney]

Remember the phrase - Photo or it didn't happen?
AI killed that. You cannot believe anything you see on screen.

 

[CXRAndy]

That guy has been on the pec deck

 

[spen666]

Remember the phrase - Photo or it didn't happen?
AI killed that. You cannot believe anything you see on screen.

Someone on radio recently said "we live in the age of misinformation" - sadly that is seemingly true

 

[C R]

Someone on radio recently said "we live in the age of misinformation" - sadly that is seemingly true

We've lived in the age of misinformation for a long time. All AI has done is increase the size of the pond of misinformation producers.

 

[Beebo]

I've dabbled with this stuff. I use Krita with the AI diffusion plug-in so my Graphics card does all the computation on my local device.

The results speak for themselves.

This one is modeled on a few key words i would describe as a brexiteer, cyclist, Heavy Keyboard user, trolling on a popular cycling forum...

View attachment 8992

The results speak for themselves.

That that down, I haven’t given you permission to use my photo.

 

[spen666]

We've lived in the age of misinformation for a long time. All AI has done is increase the size of the pond of misinformation producers.

it was much easier to identify misinformation previously.

With AI, the misinformation has become mainstream and is in cases becoming the accepted truth. Its quite worrying really

 

[PurplePenguin]

The language stuff is great. Pictures less so. It feels like being to talk to wikipedia and ask it questions. It can help with tax queries too.

 

[Psamathe]

There are ways to check being introduced C2PA (Content Credentials). Camera adds a digital signature to the image and editing software them maintains the image including records of how it has been manipulated (edit made). As it's all certificate based AI images don't have the certificates.

https://contentcredentials.org using https://contentcredentials.org/verify/

No image from camera to hand but example of one from years ago been through Adobe LrC and had edit credentials added (just to show the sort of info recorded and checked)

I have my camera add the certification to all my images and Adobe LrC maintains the certification and edit/manipulation summaries.

Sony additionally go a step further and embed 3D data from the sensor into the image and anybody questioning the photo can upload to Sony's verification site which will double check the embedded 3D data in relation to the image itself.

All driven to try and protect against AI generated images

Ian

 

[briantrumpet]

The language stuff is great. Pictures less so. It feels like being to talk to wikipedia and ask it questions. It can help with tax queries too.

Indeed - it's amazing at all sorts of things, but the stunning language output (which has moved on the discipline of computational linguistics into another league) masks its limitations. It can certainly do things that humans can't.

Amusingly, a friend of mine turned on Google's AI chat function while I was holding forth about its limitations (the absence of doubt being one of my main observations, and the role that emotion plays in human thought processes), and this utterly realistic AI voice used utterly convincing language to agree with me about its own limitations currently and for the foreseeable future.

 

[PurplePenguin]

Indeed - it's amazing at all sorts of things, but the stunning language output (which has moved on the discipline of computational linguistics into another league) masks its limitations. It can certainly do things that humans can't.

Amusingly, a friend of mine turned on Google's AI chat function while I was holding forth about its limitations (the absence of doubt being one of my main observations, and the role that emotion plays in human thought processes), and this utterly realistic AI voice used utterly convincing language to agree with me about its own limitations currently and for the foreseeable future.

Did you ever use google to search for stuff or did you stay true to the Dewey Decimal system?

 

[Psamathe]

.... It feels like being to talk to wikipedia and ask it questions....

(Using your post as an excuse for me to raise Wikipedia). Whilst people are very ready to be critical of accuracy in Wikipedia I think they mainly focus on the reliability of it being so open and overlook the monitoring and reverts, author approval mechanisms, etc. making it actually quite hard for malicious edits to stay in place for too long.

That said, from the days when NewsBiscuit was worth reading a brilliant article covering Wikipedia's 27th Birthday

Ian

 

[First Aspect]

Did you ever use google to search for stuff or did you stay true to the Dewey Decimal system?

Mr Dewey had a lot to answer for. Attempting to decimalize things expressed using language was always stupid. If it wasn't, all supermarkets would have the same layout, for example, and we couldn't have that or men would be able to use them.

AI doesn't give very hot veterinary advice yet, by the way. Basically, all advice is that if your cat is alive in some way you should get advice from a vet.

It did confirm for me that a dead cat can't get an abscess, though, which is correct.

 

[briantrumpet]

Did you ever use google to search for stuff or did you stay true to the Dewey Decimal system?

Both have their place... I used to love the slight randomness of finding stuff on library bookshelves, but also learning about how to use Google effectively (which necessitates thinking about key vocabulary, which requires some understanding of what you're looking for).

My scepticism isn't about the development of AI itself, but more about recognising its limitations, and how the use of the impressive language output masks those limitations.

Anyway, in the meantime, back to people with the wrong number of fingers and not being able to count how many esses there are in 'assess'. Cheap shots, I know, but quite fun.