I thought that openAI would train based on the dataset from all the people using chatgpt, but as far as I can tell it's never learned anything from conversations I've had with it when I tried.
It's a problem if you believe in a potential AI apocalypse.
https://www.astralcodexten.com/p/perhaps-it-is-a-bad-thing-that-the
https://www.astralcodexten.com/p/perhaps-it-is-a-bad-thing-that-the
It depends on how you define intelligence maybe. One definition is "ability to solve your problem in the shortest number of steps".
I could see that being a problem for any system that's trying to be universal.
Human beings tend to learn about and model the people around them so they can understand what to expect. If an AI model instead tries to build a generic model for interacting with a universalized average person representing "anyone" then I'd always expect it fail quite often and fall into an uncanny valley because there is no single way to participate in a group, because there are unlimited groups.
Honestly, it's kinda already happening.
Human beings tend to learn about and model the people around them so they can understand what to expect. If an AI model instead tries to build a generic model for interacting with a universalized average person representing "anyone" then I'd always expect it fail quite often and fall into an uncanny valley because there is no single way to participate in a group, because there are unlimited groups.
Honestly, it's kinda already happening.