3 ms·
Every time I ask Opus 4.5 to generate code it is a mess, by mess I mean a human would not have done it that way, it is sub-optimal for maintainability and impro
by illist-ell1s 8mo ago
Every time I ask Opus 4.5 to generate code it is a mess, by mess I mean a human would not have done it that way, it is sub-optimal for maintainability and improvements. The more complex the worse. But generally at least the code works. If I need to modify it, what a nightmare, better tell it to modify so AI slop leads to more AI slop.
It is the same as writing a story. It ain't gonna win The Hugo Award. We know it is AI, full of cliches and corporate speak. I think the difference is people are used to crap code, where as they won't read a crap book.
Most code is crap because 1) engineers don't care about the software 2) high turn over 3) getting rushed
So maybe AI just generates code at the level most people are used to, I have no idea because compared to my code Claude code is bad, but bad is not useless. I put the bad code in its little corner and it if it does its job it is fine.
- whattheheckheck 8mo agoWhat's your github profile? I really want to admire the best of the best
- KellyCriterion 8mo agoyou have to give it clear statements what you are looking for, how the solution should work and what you are expecting. also insert into prompts (or in the base prompt) someting like (e.g. example from my Claude setup) >> - Configuration as data, logic as loops - Dynamic rendering (adapts to data size/content) - Single source of truth (no scattered magic numbers) - Defensive coding (null checks, division by zero) - O(1) lookups over O(n) searches where possible