Hacker News Reader

Comment by taneq | Hacker News Reader

"Ah, but can it accurately capture the depths and intricacies of a human soul?"

"Yep."

"Yeah but a specific Appalachian human soul at around four o'clock in the afternoon on a day in mid-autumn when it looked like it would rain but then it didn't?"

"Also yep."

"Yeah but specifically at 3:56pm and the human in question is standing on loam and holding a book in their left hand and listening to Music For The Royal Fireworks by Handel?"

"Uhhh..."

"See, told you AI is useless."

View on HN

How about this conversation:

"Midjourney v5 can do hands"

"Did you look at the hands it did? There are a bunch of mis-shapen blobs, hands with extra fingers, two thumbs on either side, etc."

"Sure, but there are also some accurate hands, so it can do hands."

Does being able to do something mean you can do it perfectly 100% of the time? I'm not sure who was supposed to be unreasonable in your imaginary conversation.

"Hey check it out, this new architecture can sometimes solve X"

"HAH! Here is a counterexample where X is not solved, so it can NOT!"

If I say I can do something, especially when I say I'm "waving at the haters", what I usually mean is that I can consistently do that thing.

If I say "here's a self-driving car!" and show you a video of a car moving straight down a street and stopping at a light, would you agree that I have a self-driving car? After all, it drove itself down the street.

Can't wait for the "Select the non-deformed hands" captchas.

This is why it is important to remember to end every imaginary conversation with “and everyone clapped” right when the protagonist wins.

If it can draw accurate hands 25% of the time, then it would only take 4 tries to get it right. Seems pretty good to me

Nice! So if I want to draw 5 people with two hands a pop I can get 10 non-deformed hands a full... .0953674316e-7% of the time. I like those odds!

What point are you making here?

Reply on news.ycombinator.com