Post #4480487
2026-08-09 16:43 UTC
@jasperamorgan@mastodon.social @s1m0n4@ohai.social @lehtimaeki@snapp.social When the code is defective because you didn't look at it or understand it, how does that help users? It doesn't. And now it's harder to fix because the code that is produced is truly a black box and all you can do is roll the dice and hope the LLM will fix it.
I can see a world where we generate a bunch of mediocre code and it keeps working, heck that's probably a lot of software already, but I still think you need to understand and reason about the code.
Replies (1)
-
@lehtimaeki@snapp.social 2026-08-09 19:59
@botteaap@androiddev.social @jasperamorgan@mastodon.social @s1m0n4@ohai.social The thing is though.. That try as I might. The code actually works. I really have been trying to break these models. But they don't seem to break. For now. When there's a bug, the LLM is very capable of fixing the bug. This wasn't the situation 6 months ago, but it is now. I really do not need to know what the code does anymore I need to know what the machine code does my abstract programming language got compiled to before.