This forced claude break is definitely forcing me to deslopify my code. Claude seems to love\ using labmdas and doing long function chains where it is a pain to even know what the fuck is going on.
Its like tending to a zen garden. Yeah things are a mess, but that is just what nature does. Just observe the current state and fix what you can.
#8363
1 week ago
^Have you looked at the code it generates? Small things are okay but it is a massive slop factory at a large scale. It gets to the point where you question if there are genuine bugs or if the code it made was such a mess that it made its own bugs
#8367
1 week ago
>>8365
I have and I agree with you. The code it generates its unreadable dog shit. That means if all your code ends up as AI code you'll basically be forced to pay a subscription to maintain it.
For code I've hand written I allow it to fix bugs and periodically re-write all the slop how I like it. But for software that I've vibe coded I'll never touch it. It's crazy good at how it codes now. It does write bugs and they still get missed. But IMO: so do humans. You just have to ensure you create an obstacle cause to verify the software works. It can make that for you though there is a bit of a logical flaw here:
you're now having to trust that it correctly implements the obstacle test designed to test code that it wrote because you don't trust it. That's kind of a contradiction since if you can't trust it how do you trust that it's writing good what you say correctly? By being letting reality dictate correctness. Dog fooding helps here.
I have and I agree with you. The code it generates its unreadable dog shit. That means if all your code ends up as AI code you'll basically be forced to pay a subscription to maintain it.
For code I've hand written I allow it to fix bugs and periodically re-write all the slop how I like it. But for software that I've vibe coded I'll never touch it. It's crazy good at how it codes now. It does write bugs and they still get missed. But IMO: so do humans. You just have to ensure you create an obstacle cause to verify the software works. It can make that for you though there is a bit of a logical flaw here:
you're now having to trust that it correctly implements the obstacle test designed to test code that it wrote because you don't trust it. That's kind of a contradiction since if you can't trust it how do you trust that it's writing good what you say correctly? By being letting reality dictate correctness. Dog fooding helps here.