Elektrine lite

← Feed

@argv_minus_one@mastodon.sdf.org

2026-07-22 21:46 UTC

@borgbackup@fosstodon.org That's a non sequitur. Linters are also pretty good at finding bugs, but they aren't capable of writing code at all, let alone writing good code. It's also a non sequitur that bad human-written code somehow implies that AI-written code is better. It is entirely possible for AI code to be even worse. Finally, I'm not convinced that these stories of AI finding bugs are true. Last time I tried to use it for that, as I mentioned in , it failed miserably.

Replies (2)

  • @borgbackup@fosstodon.org 2026-07-22 22:06

    @argv_minus_one@mastodon.sdf.org Somehow I get the feeling that you need to retry and maybe get more practice with it. Maybe use a better model than in your last try. Maybe give it more context. Claude >= Opus, maybe also a recent Gemini [Pro] (High). I have the paid plan (~20 EUR/m), not sure if the free plan is as good. Maybe use Python and C code (I had quite a lot of success with that). It also worked great for html/css/js (Claude Fable + Opus). That's just my experience, initially I was also skeptical.

    Open ##4137134

  • @borgbackup@fosstodon.org In the aforementioned case of , I even prepared a minimal self-contained program demonstrating the problem. I showed the AI everything it needed to see to figure out what the problem is. Did it give me the correct answer? No. Did it tell me it wasn't sure? Also no. It just threw a half-dozen generic answers at the wall and expected me to see which one of them sticks. Although two of those generic answers turned out to be correct, I am very not impressed.

    Open ##5175196