4 ms·
I’m in the middle. I use paid models. It’s great for scripts, reminding me of commands, and surprisingly good at debugging error messages. But if I’m working in
by yoyohello13 2y ago
I’m in the middle. I use paid models. It’s great for scripts, reminding me of commands, and surprisingly good at debugging error messages. But if I’m working in our main code base I’m basically never able to just use the code it spits out. It can get me in the ballpark of what I want, but it often does weird convoluted shit on the first pass. I can keep prompting to refine the code down to something closer to what I want, but at that point it would have just been faster to write the thing myself.
I kind of like LLMs for learning new languages. Claude or ChatGPT are good for asking questions. But copilot really stunts learning for me. I find nothing really sticks in my brain when I have copilot running. I feel like I just turn my brain off, which seems kind of dangerous to me in the long run.
- sandspar 2y agoI'm concerned about the "turning the brain off" effect too. When we put backup cameras on cars it means that everyone can parallel park perfectly. Now take away the backup cameras and see what happens to our parallel parking.
- emseetech 2y agoI'm also concerned, because a lot of people are using these models for the early stage drudge-work and I hate to sound like Calvin's dad here, but I feel like those early stages are good for a developer to build a mental model of their larger goals and targets. Sure it's easy and a lot of boilerplate, but our brains are working while we're doing it, and for me, it feels like a necessary part of the process. I'm not convinced automating it away isn't going to result in a more difficult time understanding of your product and code base later on.