Post #2357572
2026-04-24 12:12 UTC
>sort and replicate shapes and tag some similar ones with the label 'butts' - you can train it to do that.
But it can never *recognise* (re-cognise) a butt. It cannot get to know butts. And it certainly can't be attracted to butts.
Which is why you get these (to a human) obvious gaffs where limbs start merging with the scenery, or there are too many of them. Because the machine doesn't have a concept of butts. It has millions upon millions of example images that it can deliver approximately>
Replies (1)
-
@rubyjones@wandering.shop 2026-04-24 12:15
>similar result too. But its success is judged on the likelihood that an arrangement or pixels will be accepted by a user as containing 'butts', not on any idea of what a butt is. It crafts nothing. It imagines nothing. It creates nothing. It just automates shifting pixels around. And the prompter applies a very tiny amount of craft composing their prompt. But even when they select a 'successful' one, that success is owed to the millions of mastercrafts people who have not been compensated.