Sweep is a really impressive demo of LLMs with coding: given your source code and an issue, it will generate a PR! Example: https://github.com/edreisMD/plugnplai/issues/65
I'd want to have a really thorough test suite if I was accepting LLM PRs though.
miniblog.
Related Posts
LLMs are surprisingly good at reducing crash samples.
I've had success with "this project crashes my static analysis tool with the following command, try to shrink the repro whilst preserving the crash. Commit each smaller version to a branch".
LLMs have been really helpful for me porting some JS projects to TS.
Since the types are erased and annotations are optional, I can go gradually and safely with more minimal human oversight.
(I'm not sure if it'll significantly help robustness, but it's lovely for navigation.)
Show, don't tell: effective advice for writing for both LLMs and people!
