hoffer

hoffer

@hoffor.bsky.social

programming/music/drawing/yapping/existing hoffer.neocities.org

38 followers32 following120 posts

i've been trying out LLMs for 'coding' and man are they stupid. 'top tier' models, dumbest things i've ever had the displeasure of interacting with. brainless. i know how they work inside and out since i've designed smaller ones to learn. every generational 'improvement' will always be a total grift

i've been keeping up with them since before they became common knowledge. the tech behind them is interesting but transformer models BY EXPLICIT DESIGN are not meant for reliability in any capacity, no matter how much 'progress' is made. NEVER trust these things no matter how convincing they sound.

for anyone who's worried LLMs will 'go rogue': that's literally not how they work at all. these corporations are desperate for you to think LLMs are 'becoming so smart they'll outpace human intelligence' but the real danger is actually how dumb they are and how many people will fall for the grift.

another slightly older project i learned a TON of math from: a simple software renderer. every single pixel was calculated manually by me in python, no opengl geometry etc. didn't follow a guide, so i had to discover how to make 2D look 3D with guesswork/intuition and math resources. 280fps at 720p!

i should post random things i make here more often.. this was a line drawing project i worked on a few months ago with the goal of intersections being mathematically lossless (no floating point error). it was intended to be the basis for a simple 3D CAD sketcher, but that'd be a massive task lol

You are up to date.