You tell an AI you want 5 of something, it looks at all the reference that it has on what “five of something” looks like, then guesses.
If we could teach these predictive engines to recognise fixed notation like numbers it would be so much easier, but the training data would still lack nuance in it’s descriptors.
How often do people talk about how many fingers are showing in what position their hand is when posting a photo? Or how many toes their chicken’s feet have?
![Peek-a-boo! [SD]](https://lemmy.world/pictrs/image/6341e9c4-c34b-48c7-aec8-bdd5d27cdc32.jpeg){kind=link}
AI can’t count, hence the problem.
You tell an AI you want 5 of something, it looks at all the reference that it has on what “five of something” looks like, then guesses.
If we could teach these predictive engines to recognise fixed notation like numbers it would be so much easier, but the training data would still lack nuance in it’s descriptors.
How often do people talk about how many fingers are showing in what position their hand is when posting a photo? Or how many toes their chicken’s feet have?