←back to thread

559 points Gricha | 1 comments | | HN request time: 0s | source
Show context
xnorswap ◴[] No.46233056[source]
Claude is really good at specific analysis, but really terrible at open-ended problems.

"Hey claude, I get this error message: <X>", and it'll often find the root cause quicker than I could.

"Hey claude, anything I could do to improve Y?", and it'll struggle beyond the basics that a linter might suggest.

It suggested enthusiastically a library for <work domain> and it was all "Recommended" about it, but when I pointed out that the library had been considered and rejected because <issue>, it understood and wrote up why that library suffered from that issue and why it was therefore unsuitable.

There's a significant blind-spot in current LLMs related to blue-sky thinking and creative problem solving. It can do structured problems very well, and it can transform unstructured data very well, but it can't deal with unstructured problems very well.

That may well change, so I don't want to embed that thought too deeply into my own priors, because the LLM space seems to evolve rapidly. I wouldn't want to find myself blind to the progress because I write it off from a class of problems.

But right now, the best way to help an LLM is have a deep understanding of the problem domain yourself, and just leverage it to do the grunt-work that you'd find boring.

replies(21): >>46233156 #>>46233163 #>>46233206 #>>46233362 #>>46233365 #>>46233406 #>>46233506 #>>46233529 #>>46233686 #>>46233981 #>>46234313 #>>46234696 #>>46234916 #>>46235210 #>>46235385 #>>46236239 #>>46236306 #>>46236829 #>>46238500 #>>46238819 #>>46240191 #
pdntspa ◴[] No.46233365[source]
That's why you treat it like a junior dev. You do the fun stuff of supervising the product, overseeing design and implementation, breaking up the work, and reviewing the outputs. It does the boring stuff of actually writing the code.

I am phenomenally productive this way, I am happier at my job, and its quality of work is extremely high as long as I occasionally have it stop and self-review it's progress against the style principles articulated in its AGENTS.md file. (As it tends to forget a lot of rules like DRY)

replies(12): >>46233446 #>>46233448 #>>46233642 #>>46233652 #>>46233782 #>>46234010 #>>46234898 #>>46235480 #>>46238997 #>>46241434 #>>46241981 #>>46242791 #
n4r9 ◴[] No.46233446[source]
I think we have different opinions on what's fun and what's boring!
replies(4): >>46233963 #>>46234282 #>>46234342 #>>46234495 #
embedding-shape ◴[] No.46234342[source]
Some people are into designing software, others like to put the design into implementation, others like cleaning up implementations yet others like making functional software faster.

There is enough work for all of us to be handsomely paid while having fun doing it :) Just find what you like, and work with others who like other stuff, and you'll get through even the worst of problems.

For me the fun comes not from the action of typing stuff with my sausage fingers and seeing characters end up on the screen, but basically everything before that and after that. So if I can make "translate what's in my head into source on disk something can run" faster, that's a win in my book, but not if the quality degrades too much, so tight control over it still not having to use my fingers to actually type.

replies(1): >>46235073 #
mkehrt ◴[] No.46235073[source]
I've found that good tab AI-based tab completion is the sweet spot for me. I am still writing code, but I don't have to type all of it if it's obvious.
replies(1): >>46237590 #
1. OkayPhysicist ◴[] No.46237590{3}[source]
This has been my approach, as well. I've got a neovim setup where I can 1) open up a new buffer, ask a question, and then copy/paste from it and 2) prompt the remainder of the line, function, or class. (the latter two are commands I run, rather than keybinds).