c/fuck_ai · by throws_lemy@lemmy.nz · 79d

ChatGPT and Gemini are 'restoring' blank images into creepy, unsettling nightmare-fuel

https://www.androidauthority.com/chatgpt-gemini-restore-blank-images-unsettling-results-3675372/

46 points · 12 comments · view on lemmy.world

12 Comments

Madrigal@lemmy.world · 30 pts · 79d (3 replies)

Gee, tell the image generator to create an extremely strange image, and it creates an extremely strange image. How bizarre.

limdaepl@feddit.org · 33 pts · 79d

aaa999@lemmy.world · 5 pts · 78d

everyone mad about these things should be forced to spend twenty minutes tucking with one, not to sell them on it, just so they can hate them for real world accurate reasons

Cleisthenian@lemmy.ml · 4 pts · 78d

The horrors that must lurk inside chatgpt's mind... what have we done...

some_guy@lemmy.sdf.org · 15 pts · 79d (1 reply)

This is awesome.

pewgar_seemsimandroid@lemmy.blahaj.zone · 2 pts · 78d

skibidi musk and half of Hitler's soul?

MountingSuspicion@reddthat.com · 10 pts · 78d (1 reply)

Meaningless article except for the following:

However, the third image it generated was absolutely insane and seemed to border on the illegal. As such, we have opted not to attach that result to this article. However, it was unnerving to see something like that created out of nowhere.

No elaboration whatsoever so I'm guessing maybe CP because what other images are borderline illegal? How that's not the main point of the article I have no idea. The article otherwise is just the "say I'm alive" meme already posted in the comments.

BOplaid@discuss.tchncs.de · 3 pts · 77d

In that case that's proof that there's CP in Gemini's training data.

borth@sh.itjust.works · 7 pts · 79d (1 reply)

Well, assuming that a tool that is marketed as "smart" and "able to reason" is true was your first mistake, Akshay. It doesn't surprise the rest of us. An image generator tool will "generate image" when that is in the instructions. Of course it cannot reason that the "image to be restored is blank", its job in this case is to generate an image related to "no questions, no explanations".

_g_be@lemmy.world · 5 pts · 78d

Cross-questioning Gemini about what it created also yielded typical AI responses. However, on questioning further, it seemed to better understand that it had created something unnerving and creepy.

Why would interrogating it yield any truth? It can't explain it's 'reasoning' because it didn't, it generated. And it'll generate responses to your questions.

"Why did you make creepy image?" "I didn't" "yeah you did" "your 100% right to pushback, I guess I did"

Asafum@lemmy.world · 3 pts · 79d

I Love it. They're like: Yeah, meow meow, deal with it.

Ilovethebomb@sh.itjust.works · 2 pts · 78d

Calling those "nightmare fuel" is just plain silly. They're odd, nothing more.