LogFAQs > #978980025

s0nicfan posted...

They're dealing with an impossible problem and trying to solve it by piecemeal patching in special rules and exceptions whenever something pops up, but that's only going to make the entire system more brittle and inconsistent. Like, that image was probably produced because they put in some sort of rule that says that any prompt involving groups of people must be racially diverse, and since there's an infinite number of contexts under which that could be embarrassing it was only a matter of time until somebody found the one that highlights how dumb that approach was. Eventually, it'll result in a lawsuit and they're just going to have to take the whole system down.

But what should bother people isn't that some public facing AI tool is having problems. It's that these same companies are developing these same tools for backend systems that will eventually do things like approve loans and what you won't have is a million people generating memes using it to show off where it fails.

On the other hand maybe an ai that collects images of people and aggregates them just inherently believes humans are racially diverse, because they literally are