It Knows What it Knows

My personal tagline (for the moment) is “telling computers what to do, they don’t always listen the first time.”

One of if not the most annoying property of ChatGPT is its insistence on turning every single response into a perfectly formed “essay” complete with introductory and conclusion/summary paragraphs.

And those are almost always filled with 💩 it thinks you want to hear. “Blah blah blah” paved the way blah blah influence seen today blah. And by that I mean it creates a statistically probable model of what it’s been told is “good.” Which is fluff and filler.

Look you fucking robot I’ve given you the instructions and context that every section I want from you is one part of a larger piece and does not need this. Does it listen? Hell no.

It’s overall great. I know what I want to see in the end. I tell it that and it usually gives me something I go “yeah ok that’s ok I guess” and I take that and tweak it myself…. or I fight with it until it finally obeys. Is it being “contrary?”

Also no. It’s just doing what it’s been trained to do, which is use its language model to spit out the most probable set of tokens it can put together. It literally doesn’t “know” any better.

Frustrated af but it’s only going to improve. Start learning this stuff I’m telling you.

#ai #chatgpt #LLM #fuuuuuuuuuuuu

I figured I’d include this meme just for the lols because its starting to seem relevant. Always be polite to your chatbot.



Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.