Amazon’s Q has ‘severe hallucinations’ and leaks confidential data in public preview, employees warn
Amazon’s Q has ‘severe hallucinations’ and leaks confidential data in public preview, employees warn
Some hallucinations could ‘potentially induce cardiac incidents in Legal,’ according to internal documents
You're viewing a single thread.
I think I kind of understand the term, but what does "hallucinations" in this context refer to? It seems like it might be fabricated unformation?
Basically, the model just makes stuff up.
Not sure why someone downvoted you. That’s exactly what the term means in this context. It’s those confidently written answers that contain false or fabricated information.
And this seems like the biggest limitation for the LLM approach. The model just knows that a certain set of tokens tends to follow another set of tokens.
It has no understanding of what the tokens represent. So it does a great job of producing sentences that look meaningful, but any actual meaning in them is purely incidental.