Or people say "I've been pumping out thousands of lines of perfectly good code by writing paragraphs and paragraphs of text explaining what I want!" its like what are you programming dog? and they will never tell you, and then you look at their github and its like a dead simple starter project.
I recently built a Brainfuck compiler and TUI debugger and I tested out a few LLM's just to see if I could get some useful output regarding a few niche and complicated issues, and it just gave me garbage that looked mildly correct. Then I'm told its because I'm not prompting hard enough... I'd rather just learn how to do it at that point. Once I solve that problem, I can solve it again in the future in .25x the time.
Here's the thing. 99% of people aren't writing compilers or debuggers, they're writing glorified CRUDs. LLM can save a lot of time for these people, just like 99% of people only use basic arithmetic operations, and MS Excel saves a lot of time for these people. It's not about solving new problems, it's about solving old and known problems very fast.
> "99% of people aren't writing compilers or debuggers"
Look, I get the hype - but I think you need to step outside a bit before saying that 99% of the software out there is glorified CRUDs...
Think about the aerospace/defense industries, autonomous vehicles, cloud computing, robotics, sophisticated mobile applications, productivity suites, UX, gaming and entertainment, banking and payment solutions, etc. Those are not small industries - and the software being built there is often highly domain-specific, has various scaling challenges, and takes years to build and qualify for "production".
Even a simple "glorified CRUD", at a certain point, will require optimizations, monitoring, logging, debugging, refactoring, security upgrades, maintenance, etc...
There's much more to tech than your weekend project "Facebook but for dogs" success story, which you built with ChatGPT in 5 minutes...
I was the driver. I told it to parse and operate on the AST, to use a plugin pattern to reduce coupling, etc. The machine did the tippy-taps for me and at a much faster rate than I could ever dream of typing!
It’s all in a Claude Project and can easily and reliably create new modules for bash commands because it has the full scope of the system in context and a ginormous amount of bash commands and TypeScript in the training corpus.
One good use case is unit tests, since they can be trivial while at the same time being cumbersome to make. I could give the LLM code for React components, and it would make the tests and setup all the mocks which is the most annoying part. Although making "all the tests" will typically involve asking the LLM again to think of more edge cases and be sure to cover everything.
Yea, obviously not, but the smaller problems this bigger project was composed of were things that you could see anywhere. I made heavy use of string manipulation that could be generally applied to basically anything
Really? Come on. You think trying to make it solve "niche and complicated issues" for a Brainfuck compiler is reasonable? I can't take this seriously. Do you know what most developer jobs entail?
I never need to type paragraphs to get the output I want. I don't even bother with correct grammar or spelling. If I need code for x crud web app who is going to type it faster, me or the LLM? This is really not hard to understand.
I recently built a Brainfuck compiler and TUI debugger and I tested out a few LLM's just to see if I could get some useful output regarding a few niche and complicated issues, and it just gave me garbage that looked mildly correct. Then I'm told its because I'm not prompting hard enough... I'd rather just learn how to do it at that point. Once I solve that problem, I can solve it again in the future in .25x the time.