Have one of those I actually thought was kind of funny, and bit like having a conversation with an AI.
Tried searching for quotes from the Matrix because of all these AI issues and asked:
"Quote of Agent Smith to Mr. Neo 'How will you speak with no mouth?'"
and got back:
"You're wrong. Agent Smith never refers to Neo as anything other than Mr. Anderson." Completely did not even try to answer the question.
Course, these days it's more Harlan Ellison "I Have No Mouth, and I Must Scream" [1]. Kind of went beyond the Matrix to total thought control and constant machine torture. "How will you search if all results are false?"
I had Google AI unambiguously assert that a passive LC network could never have an output voltage higher than the input voltage because that would need an active device. It's just dangerous.
How do you think it will raise the voltage if it just seats there, being _passive_ ? You need an _active_ network to raise the voltage. /s
20 years ago in the "AI" class, the proffesor said, that the biggest problem with (machine) learning, is that sometines, it will learn wrong, and you have to teach it again.
And nowadays it sometimes just silently drops a huge range of matching verbatim results without even informing the user.
It’s especially serious when searching for UN related documents, which have a special document symbol structure.
e.g. 99.9%+ of the documents under the classification A/C.1/… simply don’t show up with a “A/C.1/“ search, except for a handful. And it’s not like irrelevant results are clogging it up either, as Google only returns 40ish results for the whole internet. When there’s thousands of publicly accessible UN documents in that category…
Me: “exactly-this-thing.py”
Google: You misspelled “sorta-related.js”. Here you go.
Me: Did I stutter?!