Reply to post: Re: We're doomed

Russian criminals can't wait to hop over OpenAI's fence, use ChatGPT for evil

Zolko Silver badge

Re: We're doomed

Imagine an AI that learns from its mistakes

I've read an interesting analysis about this (may-be here on ElReg in a comment) that suggested that since the current AI learns from the Internet, and then outputs on the Internet, it might feed its learning mechanism on itself, creating a positive feedback loop, which is unstable. So all this hype is short-lived and doomed to fail : it will sort-of work only as long as humans – involuntarily – feed it.

Which doesn't mean it can't bring havoc during the time of its – useless – existence

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon