I think that just like humans, AI will make logical mistakes at some rate linked to lines of code.

Like I remember reading human studies that people make 1.5 - 5 errors per 100 loc.

If AI works in a similar way then we should stick to higher level languages that minimize loc