Re: Artificial Intelligence, the hobby and the USRA
This was quite possibly the last place I expected to encounter OpenAI and ChatGPT discussion.
Some friends of mine as well as myself have been experimenting with these ChatGPT-based generative algorithms over the last several months, mostly to create helpful or interesting bots for online discussion servers (chat rooms). One thing we've discovered is that the responses they return have the semblance of truth without being true. Ex: when asked what day it is, the algorithm will return a value that was included in its original training data, often years in the past and usually mixing up days of the week with the actual calendar day. ("Today is Monday, August 29th, 1997".) Mostly this is because these algorithms were generated from and trained on real and factual sources; sources which they will regurgitate plagialristicly exactly as a confused and failing student will plagiarize material for their own papers without the slightest bit of understanding. There is some truth in what it returns, but it is a truth in which the AI has no ability to determine the real from the fantasy, and so freely mixes both.
It very much reminds me of the shopkeeper in Philip K Dick's _The Man in the High Castle_, who noted that the "antique" items he sold did not need to be actual historical relics, they merely needed to have enough of a historicity attached to them that someone would buy them.
Don't buy what the AI sells you.
That said, it's about as reliable as most of our long-term memories. I've had people tell me things with absolute conviction that I know to be otherwise false, such as someone who insisted the Toledo Peoria and Western was a freelance model railroad and never existed in real life, or that there was only one place in the entire state of West Virginia where two railroads crossed at grade. I've probably even heard people say that all double-sheathed boxcars were insulated. And I'm certain many of us have encountered railroad history books in which captions can be verified as false by simply looking at the picture they caption. I've no doubt unknowingly repeated some of these true-isms.
In that, the AI inherits the fallibility and error-prone nature of its creators.