Why does AI have such a problem with fingers?

It does not understand jokes either.
I told it one of Jimmy Carr’s latest jokes: “Climate change is like my niece. Getting hotter every year”.
And it did not understand it at all.

Not just fingers. Necks, feet, any complex movement. Student of mine used AI to create some video clips for a project. They were creepy as hell. It’s that uncanny valley problem where things appear real but unreal at the same time. No nuance — yet.

Give it time.

Lack of training data and the various complex configurations that hands come in.

Don’t worry. They’re getting better at it and it’ll be solved in time.

It bothers me.

It should bother you. It uses creepy algorithms to somehow create images based on statistical shuffling of other images and databases created with many many hours of dirt-cheap or appropriated labor, to synthesize new images based on data associations, that often look very compelling and like something that was created by a human artist, or a camera, and yet, isn’t actually done the way a human or camera would do it, but based on weird poorly-understood data manipulations that don’t really understand what they’re doing or showing.

It’s mostly fixed now

Some models are also better than others for different tasks.
I have had great results with Pony Diffusion and its various offshoots, though depending on the settings you use, issues will creep in (fused body parts, floatinng body parts, duplicate body parts, etc)

@Dutchess_III you might regret how much AI has learned the next time you give an ATM the finger and it decides you don’t really need that cash.