I know posting your AI chats is a bit passe already but I'm not sure I've seen anything else this simple to, like, "prove" to myself that there isn't any sort of "intelligence" involved in this stuff.
Also goddamn does "HR Neutral" as a default tone grate on me quickly
The "best" thing about this is that eventually a human responsible for the bot will notice how embarrassing this line of questioning is and likely add a special case answer and Some People will say "see? It's improving every version, I think there really is something here"
Artificial "intelligence" really is just the programmers manually writing out every single possibility the model could interact with - trial-and-error so everyone sees it fail first - and then pretending the machine figured it out on its own.
a little bit of this due to the human-mediated "reinforcement learning" section, and a lot of, well... the fact GPT models are basically markov chain generators with a neural network bolted on as a compression routine. they get fed a prompt, then tack tokens onto the end of that prompt repeatedly until either their next token says to stop, or some filter forces them to stop
it can't even tell the difference between things it said and things you said, in the raw, so of course it can't tell how many words it's going to say back to you!
I've been real annoyed by how much hype is surrounding these things right now (apparently over $10 billion worth!?) so this is a real good demo