Wanted to test how far guided LLM coding could go with a real, practical, shippable weekend project.
-
Wanted to test how far guided LLM coding could go with a real, practical, shippable weekend project. The answer: a two day project done in an hour and a half.
Aside from picking content, I wrote 0 LoC for this project, including readme and CI. The tech has gotten alarmingly effective in the last few months. Might blog about it soon.
The project: A web page that plays a random "work/study music" YouTube video, with basic filtering for various games/moods. https://stevestreza.com/lockin/
-
This isn't an endorsement of LLM driven coding. But as I've said all year, it's important to understand the state-of-the-art.
-
@stevestreza out of curiosity, which one did you choose? i’ve found Claude to give ‘better’ results than others given the same prompt… but I still wouldn’t be able to necessarily spot bad code unless it just straight failed to work.
-
@stevestreza Also I read LoC as Library of Congress and got very confused until I switched gears.
-
@octothorpe This was built with Cursor, which uses claude-3.5-sonnet. Would like to try other models, especially local ones, to do something like this.
-
@stevestreza Yeah. Curious as to your findings.
