OpenAI: AI hallucinations are mathematically inevitable, not engineering flaws(computerworld.com)

OpenAI: AI hallucinations are mathematically inevitable, not engineering flaws

https://www.computerworld.com/article/4059383/openai-admits-ai-hallucinations-are-mathematically-inevitable-not-just-engineering-flaws.html

4 comments

signa1110 mesi fa

this is imho pretty cool for humans :o) if hallucination is inevitable, then the whole thing cannot function without a-human-in-loop.

this ofcourse is in direct confrontation with fundamental premise of adopting such tech to reduce human workforce.

ctxc10 mesi fa

Pardon my French, but...no shit?

brettkromkamp10 mesi fa

This! Exactly, this! As if this is some kind of novel revelation.

dietr1ch10 mesi fa

It's a problem in the underlying model, not an implementation flaw, but still an engineering flaw in my book.