- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
Not the author, but I’m interested in discussion about how you’re using AI.
I’m not as stuck on Claude as a model as much as the author. I find their limits on rates really too low to use as an effective coding partner, I spend more time waiting for it to time back in than actually doing something useful, GPT4 does better with less hanging. O1 is better, and I haven’t figured out how to use Deepseek on Cline yet. I’m not going to use a different editor than VSCode to code in, so Cursor isn’t really interesting.
I don’t use AI for much other than that, as I find the search results on things like Perplexity kinda worthless compared to what I can come up with without trying very hard. And chatting with an AI isn’t something I’ve found useful either.
I like bouncing off ideas. And sometimes I do role-play/worldbuilding/storywriting with it. Like a crude version of a dungeon master in a D&D campaign. I mainly use KoboldCPP for that and a local model. And the occasional “give me 5 ideas about …” or I copy and paste a text in a foreign language and have it translate it. I don’t think I have much use for AI apart from that. I’ve tried programming, summarizing, rewriting emails, and it doesn’t really help me for what I’m trying to achieve. At least concerning large language models. That’s only one variant of AI. I also use AI algorithms in my phone camera, occasionally to transcribe audio, my email spam filter likely uses some form of machine learning… I’m tinkering with software like Immich and hope it can classify and sort my pictures…
(I wouldn’t use it to cross-check my finances, though. I found AI is particularly bad at arithmetic. It’ll fail with the first proper division in a calculation. And I’m not sure about the code review. I’ll have to try. But most bugs I introduce are more because of some edge cases and I doubt you can tell from a diff. It’d need a solid understanding of the whole codebase. I think it’s wiser to spend the time it’d take you to read an AI code review on doing some unit tests or running the code once.)