All hyperbole aside, an LLM is just an algorithm for calculating the statistically most likely thing that goes in the blanks of an incomplete text. It's not nothing, and it's certainly impressive, but it cannot do *any* of the things we're told it can do.
Exactly. I’m running various LLMs locally and “simulated intelligence” would be a better descriptor, imo. Like any other tool, it can be used for good or ill.
The only issue with "simulated intelligence" as a phrase is we're dealing with people who think a good enough simulation of the thing is the thing, i.e. the "Less Wrong" crowd.
My original version of this said "figuring out" instead of "calculating." I wanted to make it clear I did not intent to imply the machine does any thinking, whatsoever.