You write the way I’d imagine someone who has never used a modern LLM to write. You are correct that they don’t reason, but your dismissive tone is unwarranted. They are extremely useful tools when directed by someone who knows what they need. It’s a tool, and like any tool, it’s only as good as the one wielding it. That doesn’t mean the tool is inherently useless.
No one here was suggesting that LLMs will replace developers. They are suggesting that LLMs will replace compilers.
Just so.
Anyone claiming that “LLMs don’t reason” at this point simply hasn’t used a modern LLM. Claude is functionally indistinguishable from having a human employee working for me. It’s astoundingly good.
Whether it’s “reasoning” in the strict sense of the word that we reserve for human brains is immaterial. That’s academic nonsense for philosophy undergrads to argue about. Watch it in action and try to tell me it’s just a dumb token predictor.
The world is changing.
And yet, even stochastic parrots can be useful: Replacing the Type Checker - #139 by JanWillemBrands
Okay, this thread is definitely over.