Fuck AI

Something weird that I noticed about that: LLMs are doing exactly the same thing in French. "It's not, it's" becomes "ce n'est pas, mais c'est" (in English: "it is not, but it is")

It's kinda improbable that the same sentence would happen in both languages, but there it is.

My personal pet peeve is everything having a 'why it matters' section. Not everything needs to matter, it can just be fun. I already clicked the link, no need to convince me. Also it's often just restating the opening paragraph. And it makes everything sound like a marketing piece.

That "it’s not x, it’s y" - thing is really weird. I think it's even discouraged to talk like that for humans, as it frequently gets people confused. And if the audience is not paying attention 100% they might as well memorize X instead of Y. It's also mostly unnecessary. If something is Y, just say so. (There's a few limited use-cases. Like when debunking proper myths. Or in an Ikea instruction leaflet with the wrong screw crossed out and the correct one depicted next to it. And other textbook examples like that. But it shouldn't be a normal speech pattern, by any means.)

I had ChatGPT do other weird things. I tried writing Python code with ChatGPT. And on more than one occasion it gave me a full screen of text, or computer code... Which I read and copy-pasted... Just to scroll down and find out it goes on to say ChatGPT does NOT recommend doing it that way, and instead do...

And that really made me feel mocked. Why waste 3mins of my life then, by making me read and copy-paste the wrong "solution"? I don't know why that is, though. Maybe it came to some "realization" while writing it, and since it's writing left to right, it can't make corrections. So it'll decide to weasel itself out and portray it as some "lesson" in what not to do.

Feels somehow like the "it’s not x, it’s y". Just with the "not" coming after explaining X. And everything is stretched out over 5 pages.

Or maybe it'a just because they made it prefer longer answers, or give several options/perspectives... And adding wrong ones contributes to length.

The anthropomorphism itself goes with this also.

Oh and how they’re occasionally like “you’re not imagining things, …”, like bro, I just told you that, you don’t need to be condescending about it.

A core training principle in any LLM is to be helpful. This tell is a creative way to self-validate a claim, even if it is weak. A common goal that is difficult to do is to retrain an LLM to prioritize being factually accurate vs. always finding some answer that makes the user happy. The reward system for that behavior is very deep and hard to break. AI could have been... rather was in the ANI forms, a decent tool until someone figured out how to make profit from hype, and now the damage is done.

Training material is one Thing. Setting Up Something with the LLM the other.

You gotta Set guardrails and logics especially when it becomes Something more detailed. In the right Hands, absolutely fine. But if you want to manipulate or whatever..yeah. They will do for the sake of money in the one or the other way.