Ask HN: Any great ML code reviewer tools?
It seems like a huge productivity opportunity compared to coding assistants (copilot, etc). However I'm not seeing equivalent investments/tools from companies.
Are there any great tools for ML code-reviews that integrate into Github?
I think this should be as big or bigger than coding assistants, and could (once well tuned) be more helpful. Lots of good reasons:
- UX: Github/Gitlab already have great CR tools. There’s no need to invent a new UX paradigms; just use existing ones with existing APIs.
- Long context LLMs: can more of the codebase into the context, even all of it for smaller projects.
- Efficiency: LLMs with large project-size context take a lot of memory. Unlike coding assistants which run for long periods, CRs can run as a job and use lots of very large contexts but only momentarily.
- Value: finding bugs before they happen, and saving my team members time for first-pass code reviews.
Of course this doesn't replace humans (yet). However code reviewing is tedious and slow. Finding issues like typos and using ">" when I needed ">=" are all good tasks for ML, and make a good first pass before bringing humans in the loop.