My hypothesis for this is:
1. React allows to do the "bad" thing pretty easily, so much there's an entire page in the docs on how NOT to use useEffect.
2. Most of React code on the internet (the training set for LLMs) do the "bad" thing.
The combination of these two leads to awful code.
Genuinely curious about "Built features in Rust for the launch (I donโt even know Rust), 90% AI-written"
- How do you build features in a language you don't know?
- If you don't know the language how do you know the code is good and not slop?
- The remaining 10% is done by someone else that knows Rust?
Just released the Virtuoso data table component. Row and column virtualization, @shadcn wrappers, local and remote data model, state persistence, and the usual column management features.
https://t.co/2IDEX3D0r2
The sweet spot for me is to tell the LLM to not write any code, rather give me the output and where to paste it.
This works for me for two reasons:
1. I can spot any mistakes it has made and correct course early on
2. Way easier to review as I paste, than to do a final review of a huge amount of code
not having to type is really nice, but i think i want to go back to manually writing the code myself and more leveraging LLMs for research and understanding of the codebase
it's just too easy to defer you thinking today and end up in a bad state
we have a tedious task of moving off of bun file apis
we had opencode analyze all usage and come up with a plan for how things should be mapped
and now it's running in a loop opening one PR per file - each of which can be reviewed and safely merged independently