Skip to content

Comment on Ask HN: Is there a metric for AI code quality?

Comments

Not sure if I got the question right, but there are benchmarks like SWE pro and stuff. There's whole another debate whether you can trust it or not, and whether the labs are training on those benchmarks, but that's one way to measure that.

Other than benchmarks, I'd say that's your own test suite

i would never trust benchmarks tbh most of the new model releases do benchmaxxing

Sad, but fair!

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.