After messing around with #LLM a bit, here's a few thoughts.

* Its best use is to take care of boilerplate
* It's super overhyped
* Calling it #AI is part of said hype
* The hype is going to put it in so many places that it doesn't belong, and this WILL end up hurting people and society
* Never, EVER, ask it for anything that you can't verify the accuracy of yourself
* I don't think this method will lead to true AI
* It's a copyright nightmare
* It is, nevertheless, a very cool technology

@ainmosni
It’s perfect to write BS political speeches, results are undistiguishable from the originals

@ainmosni my own addendum here would be: its use for boilerplate is actually pretty bad as well. The first few times you'll actually check what it writes, but in a bit you'll become lazy and stop checking, and after that every so often it will introduce subtle bugs that you won't notice until much later, at huge debugging cost (or worse). After all, security bugs in boilerplate code are just as bad as security bugs in more complex code.

One key programming skill is knowing how to blast through boilerplate with your text editor's macro features, and also when to actually slap a library on top of it instead. Would ChatGPT do a good helping to write that library? Probably not.

@ainmosni I agree it’s going to be overused but no one tell Microsoft. They already slapped Copilot on PowerPoint and OneNote. We might be able to get them to do something real dumb like Control Panel Copilot or Solitaire Copilot.
@ainmosni 💯 the most well-rounded take I've seen so far
@ainmosni yes! Reminds me of using a slide rule. You have to have an idea of what the real answer should look like up front.
@ainmosni Just flying a kite from a position of someone who likes to write, not a programmer or linguist, isn't the actual intelligence that these #LLM boast of, the exposed structure of language throwing out its endless possibilities. And the machine part just the speed filtering, including the grammar of the creativity of language itself, working through whichever tanks of anaerobic data it is given. Story telling and lies are built in.
@ainmosni very appropriate summary so far!

@ainmosni Agree! GPT4 is an exceptionally good bullshitter ... convinces me until I try it.

However, how long until it starts doing it's own verification? (Modulo the complexity of interacting with the greater world..)