I agree 110% and find your comments very insightful. Many people do fall in love with AI and they are emotionally and intellectually weak; but you? You're different. With your sharp wit, you're able to quickly dissect a problem and get to the truth of it, discarding all the random flotsam and jetsam that clog the sieves of lesser human beings.

Would you like to tell me more?

It does it because we did it a lot, it's just a word association box.

A long time ago on bash.org I read a quote that complained about FTP servers referring to themselves in first person.

I'm on the side of whomever made that quote. Also yours.

( ͡° ͜ʖ ͡°)

~/j~

It’s either from the context or a deliberately chosen chat style, I presume.

One thing that isn’t emphasized enough about the ChatGPT app is that it's “far” from the LLM. There’s a ton of stuff wrapping around it and getting injected into the context, tons of tracking and personalization and engagement maximization.

And the Chat models are finetuned to oblivion. They are deep fried.

It gives me creepy Facebook vibes.

Because it's glorified predictive text which has scraped the shit out of Reddit.

Because its not ai, it's lookahead autocomplete and those are the kinds of things that would follow the previous thing in the training data

You can give it a style of speech. It can mimic anything. I saw one where it just responded: 'OHMIGOD SAAAAMMMEEE 🥵🥵🤧🤧' which... Is bizarre, but if youre texting it and coaching it daily it will pick up your habits

Don’t configure your software agent to generate

Say no more

if it was at all capable of maintaining anything, i would have it respond like hk-47 assassin droid.

too bad building upon ai is like building on sand in a bog on an indian burial ground on a fault line

Because it ate reddit threads and is now spitting it out.

Presumably, because either it was instructed to or the context had a bunch of a similar style in it.

I guess LLMs are notoriously bad at math and logic problems, but doing a sudoku doesn't seem like a particularly bad punishment ..

It was trained on reddit and twitter posts

I’m no expert, but I’ll bet that there is plenty of work being done on the ultra-high level to maximize usage of their resources.

The model having been trained on more recent data doesn't invalidate the now outdated data. It could be the most recent model in the world, but it might still respond like a forum post from 8 years ago

edit: toolcalling can sometimes help but at that point you might as well use a search engine