[mrtksn]

LLMs are quite accurate for programming, these days they almost always create a code that will compile without errors and errors are almost always fixable by feeding the error into the LLM. I would say this is extremely precise text generation, much better than most humans.

Just like with CNC though, you need to feed it with the correct instructions. It's still on you for the machined output to do the expected thing. CNC's are also not perfect and their operators need to know the intricacies of machining.

[gitgud]

> Just like with CNC though, you need to feed it with the correct instructions.

CNC relies on precise formal languages like G-code, whereas an LLM relies on the imprecise natural languages

[co_king_3]

> LLMs are quite accurate for programming, these days they almost always create a code that will compile without errors and errors are almost always fixable by feeding the error into the LLM.

What domains do you work in? This description does not match my experience whatsoever.

[mrtksn]

I'm primarily into mobiles apps these days but using the LLMs I'm able to write software in languages that I don't know with tech that I don't understand well(like bluetooth).

What did you try to do and the LLM failed you?

[RAMJAC]

The guy is a troll. LLM coding posts are flame bait now.

[co_king_3]

I'm not trolling. You're peeved that you don't have a rebuttal.

[RAMJAC]

Who said I was peeved. You are trolling. We all have a better use of our time than this.

[co_king_3]

I am not trolling.

Why are you putting words in my mouth?

I suspect it's because my genuinely held opinion offends your precious sensibilities.

In other words, you're mad.

[fendy3002]

I found that LLMs are quite accurate given a proper pseudo-code & not using library. For business-logic or vague instruction they're bad.

Still not as accurate as CNC machine, maybe early model typewriter?.