Ancestors

Written by rayslava on 2025-01-27 at 13:50

During the weekend I've been fighting with ChatGPT and Claude in order to create a simple TUI using ratatui for "windows" and sm for keeping state. And neither of them made something useful, even both of them correcting one another could not keep the logic construction big enough to track the state, update it appropriately and reflect the transitions on the screen.

At last I tired and wrote it manually.

So #LLM is a good tool but not a universal one even in the hype field. Guess, it's still a little bit early for retirement if you do something more complex than generic web-service.

However the boilerplate code and tests were created fast and without much effort (not tests for the state machine itself though), so the tools are useful and help you to get rid of some routine.

TLDR: we still have the most weird programming for us, my fellow developers!

=> More informations about this toot | More toots from rayslava@lor.sh

Written by 41402 on 2025-01-27 at 13:54

@rayslava have you tried that new deepseek?

My favorite test with them is sending them decompiled code and asking to explain what does it do. So far, it at least figured out what it does.

=> More informations about this toot | More toots from a1ba@suya.place

Written by rayslava on 2025-01-27 at 14:11

@a1ba tried an 14B version locally—it worked much better than codellamas for simple cases. Do they have a cloud version with large model too?

=> More informations about this toot | More toots from rayslava@lor.sh

Toot

Written by Alexey Skobkin on 2025-01-27 at 15:09

@rayslava @a1ba

It's no surprise since Codellama is based on LLaMA 2. It's so old now. Even LLaMA 3.0 looks bad if compared to LLaMA 3.1 or 3.3 🤷‍♂️

=> More informations about this toot | More toots from skobkin@lor.sh

Descendants