This is the gap that LLMs have been filling quite well now in my experience. There are these little isolated tasks that are described easily in natural language for which the idiomatic code snippet can be created and importantly modified by LLMs as needed. No need to pull in pesky dependencies. Though when to switch over to a full blown dependency is still a judgement call.

▲

2muchcoffeeman 7 hours ago | parent [-]

This is such a weird comment to me. These are trivial functions that could be faster to code than describe. I’d hope people aren’t relying on AI for functions as simple as clamp.

I mean, people have been writing these simple functions over and over for decades when they just needed one or two things that importing a library wasn’t needed. I wasn’t aware there was a gap to be filled.

	▲	lenkite 2 hours ago \| parent \| next [-]
		Man, people are prompting and copying LLM output for min/max functions. I saw my junior do this in a meeting - it was apparently too much work to lookup the standard library. When most people work with a LLM, the thinking part of their brain gets switched off for some reason.
	▲	wavemode 5 hours ago \| parent \| prev \| next [-]
		Clamp is a simple enough function that if you simply typed its name and arguments, copilot would autocomplete its body. No prompting needed, in that case.
	▲	hawk_ 5 hours ago \| parent \| prev [-]
		Not clamp per se, but there are enough tiny snippets which can have off by one or other issues not relevant to when you think about the problem description.