Balanced opinion on LLMs - 90% of the code for this app was LLM written. -...
Balanced opinion on LLMs
- 90% of the code for this app was LLM written.
- Maybe 5% of the feature ideas were LLM.
While it understands ideas if you tell them of your idea, LLMs (for me with this app), are lousy at thinking up useful features.
Self-replies
I see LLM generated apps on pypi and I've given a few of them a full look over. Some of them are
- Rehashes of already existing apps, i.e. the vibe coder didn't give the LLM any ideas to work with.
- Poorly focused. LLMs make ideas like a 3 year old. "Let's make a car we'll add ninjas! and and dinosaurs! and and and popcorn! and machine guns! and grandma!"
Anyhow, so far the projects I've completed have been useful to me, but wouldn't have been if I'd relied on the LLM for the ideas.
This app is a good example (I didn't write it)
It is a bunch of different things. I use it for the GH Actions version updater. Everything else is just so far out there. And the GH Action updater- I probably should find a better one, as this is not a novel idea.