Post #4323518

Even assuming that clanker-based code review has an acceptable false positive rate (citation needed on that one), the UX is just absolutely horrible. I can run the relevant linters locally, and immediately fix the issue or add a //nolint:whatever , while for llms there is no nolint tag, and i have to send a v1, wait for it to take it's sweet time to review, and then send a v2, while annoying everyone. Even if i hypothetically would scrape every single prompt they use and run the same models - the thing is nondeterministic and i can't be sure that whatever i send will pass the "linters".