3 ms·
As if humans always produce high quality always right working code without tech debt. The article is full of AI cliche echos.
by zihotki 1y ago
As if humans always produce high quality always right working code without tech debt. The article is full of AI cliche echos.
- Capricorn2481 1y agoI've never had a colleague make up a method that doesn't exist and then hand it to me saying it was done. That's not tech debt, it's wrong in a way that's distinct from humans. The problem with LLMs is they treat all context as significant. Even telling it to ignore previous context is just confusing it more. So Cursor gets hung up on dumb shit like insignificant linter errors or insisting that extracting a block of code to a function will fix a bug (what?). You add more cursor rules to keep it from doing these things, which doesn't focus it, it just pollutes the context more. It doesn't know when to ignore files it's already read. It just gets shittier and shittier. I appreciate that I can tell someone where the problem is and they'll know what is or isn't irrelevant. Edit: Here is something Claude Sonnet 4 told me today. I would expect this if I was talking to it for a while, not the second request. "Used excludeSwitches (plural) instead of excludeSwitches"
- qwertylicious 1y agoYeah, LLMs share a lot of the same challenges as self-driving cars: when they work great, we get complacent, and then when they fail, they fail in ways that humans are really bad at anticipating. And the discourse around all this is also the same: the detractors point out these flaws, and the proponents chime in with "Yeah well humans suck too" and around and around we go...
- Quarrelsome 1y agoI can teach a human, teaching an AI is much harder. It has no emotional hooks you can put things on. It has no shame.