Art is in the AI of the beholder

Just received the photos of a sewer scope inspection for a client and the AI series of description fails gave me a giggle

53 points · 10 comments · view on lemmy.world

10 Comments

cordlesslamp@lemmy.today · 6 pts · 2y (4 replies)

At least they got #3 and #9 kinda right, #3 and #4 somewhat right.

MudSkipperKisser@lemmy.world · 2 pts · 2y (3 replies)

I mean, I definitely can’t argue with “closeup of a dirty surface”

cordlesslamp@lemmy.today · 1 pts · 2y (2 replies)

Honestly, I'm surprise non of them are guessed as "colon" or "colonoscopy".

That's my initial thoughts at the first glance before reading any text.

dingus@lemmy.world · 2 pts · 2y (1 reply)

If these were colonoscopy pics, someone would be having a huge problem with their colon

intensely_human@lemm.ee · 1 pts · 2y

Sir we’ve located what appears to be a camera in your butt

Rhynoplaz@lemmy.world · 6 pts · 2y (1 reply)

To be fair. That one really does look like a Siamese cat

MudSkipperKisser@lemmy.world · 1 pts · 2y

I see a uvula more than anything on that one

XeroxCool@lemmy.world · 3 pts · 2y (1 reply)

BRB, taking a close up of a black HOOOOOooooooooo.....

Rhynoplaz@lemmy.world · 1 pts · 2y

Who you calling a black hoe?!

scrion@lemmy.world · 1 pts · 2y

None of the training data for the commonly available AI models probably includes a significant amount of footage from endoscopic pipe inspection cameras, so no surprise here.

That being said, it's absolutely possible to feed those images to a pre-trained model to continue its training and get good results.