yeah this is another thing that pisses me off about ai. Living things hallucinate. AI lies.
it doesn’t have the ability to hallucinate, it lacks the wetware. they want to use that label though because it’s humanizing, and helps to cover up their incompetent lying chatbot behavior. hallucinate is also very fuzzy - who’s to blame, the shitware mfgr, the prompt dolt, the training corpus, the inference assholes, who knows… one of many garbage terms shanghai’d by LLMs to make them seem less incompetent and more human-like.
LLMs don’t hallucinate, they lie when they can’t reach a verifiable answer. And then suckers pay money for it.
I say it’s lying because it doesn’t simply offer up one incorrect answer, it’ll keep feeding more after the first error.
if it stopped at one, I’d call it an error. the fact that they keep pushing more incorrect garbage out after the first failure to me seems deliberate and malicious.
Look, living things lie. A rock doesn’t lie and doesn’t hallucinate. You say it’s stupid to say it hallucinated (which I agree with) but then you do the exact same behavior with lying just for opposite reasons. I get it, saying it lies makes it seem malicious and you hate it and all that. In the end though, it’s an inert tool, it’s an advanced auto complete. It isn’t conscious, it doesn’t lie, it doesn’t hallucinate.
I get it, saying it lies makes it seem malicious and you hate it and all that. In the end though, it’s an inert tool, it’s an advanced auto complete. It isn’t conscious, it doesn’t lie, it doesn’t hallucinate.
it’s malicious that the vendors portray it as hallucination.
LLMs are not AI.
Calling it an inert tool at this point would be shortsighted, it is an advanced autocomplete, and I tend to roll dario/altman’s ‘excessively inaccurate portrayal’ of these failures as hallucination as a lie.
yeah this is another thing that pisses me off about ai. Living things hallucinate. AI lies.
it doesn’t have the ability to hallucinate, it lacks the wetware. they want to use that label though because it’s humanizing, and helps to cover up their incompetent lying chatbot behavior. hallucinate is also very fuzzy - who’s to blame, the shitware mfgr, the prompt dolt, the training corpus, the inference assholes, who knows… one of many garbage terms shanghai’d by LLMs to make them seem less incompetent and more human-like.
LLMs don’t hallucinate, they lie when they can’t reach a verifiable answer. And then suckers pay money for it.
Saying it’s lying is even more anthropomorphic. If people were honest (on both sides), they would just call it software error.
I say it’s lying because it doesn’t simply offer up one incorrect answer, it’ll keep feeding more after the first error.
if it stopped at one, I’d call it an error. the fact that they keep pushing more incorrect garbage out after the first failure to me seems deliberate and malicious.
Look, living things lie. A rock doesn’t lie and doesn’t hallucinate. You say it’s stupid to say it hallucinated (which I agree with) but then you do the exact same behavior with lying just for opposite reasons. I get it, saying it lies makes it seem malicious and you hate it and all that. In the end though, it’s an inert tool, it’s an advanced auto complete. It isn’t conscious, it doesn’t lie, it doesn’t hallucinate.
In any case, I was just trying to be clever.
it’s malicious that the vendors portray it as hallucination.
LLMs are not AI.
Calling it an inert tool at this point would be shortsighted, it is an advanced autocomplete, and I tend to roll dario/altman’s ‘excessively inaccurate portrayal’ of these failures as hallucination as a lie.
in any case, it’s all garbage.