Post

It's easy to keep nudging the AI until it 'fixes' the tests, and it feels like achieving something, while the value of such tests may be plainly negative. Seeing too many false positives like that lately.

23/9/25, 8:25 am0 reposts2 likes

Paweł Świątkowski

@katafrakt.eurosky.social

23 Sept 2025

100% this. I have experienced many times that if you push an LLM really hard on writing a non-trivial test, it will at some point just end up with assert 1 = 1 or something on the same level of creativity.