Hacker News Reader

Comment by wobblyasp | Hacker News Reader

I'm confused.

Assistants are so easy to use yet we should drop everything to "learn"/"adopt" them? Why not just wait until they're not complete dumpster fires then start using them?

View on HN

I haven’t looked into these coding LLMs, but I wonder—they’ll hit diminishing returns or the limit of the model at some point. It might be before they are just, like fully sentient fellow developers that we can describe our wishes to in vague terms to get working code.

Maybe it is far enough along that we can at least see what the final form would look like and start getting comfortable with working inside those constraints? I don’t know.

Some time spent now learning how they perform will mean you're better prepared when they're ready.

What are you "learning"? "Prompt engineering" is hardly a skill and varies LLM to LLM.

*if they become ready

Yep.

> This wasn't something that existed, it wasn't regurgitating knowledge from Stackoverflow. It was inventing/creating something new.

Except it was. It wasn't regurgitating the exact answer it gave you, but it was using the collective knowledge of StackOverflow/other code to generate the solution.

Give your head a shake.

To me, "using collective knowledge" is not mutually exclusive with writing novel code.

Fundamentally I only know about a language's functions/keywords/types/... because I've read them in some documentation or code snippet somewhere, but if I combine them in a new way to solve some previously-unseen task, I'd consider that to be creating something new.

Of all the (valid) downsides of LLMs, this is maybe the weakest one. I'm not even sure what would qualify as "new" for you.

Anything I've ever done at work could have been pieced together from StackOverflow. Anything you have ever done can most probably be pieced together from StackOverflow as well.

I sincerely doubt it; but I also work in an Org that has highly internal code with none to poor documentation.

Also, I never said it was a "downside". I'm simply stating that the author is exaggerating what the LLM is doing. It isn't creating novel code.

I used LLMs all the time to spit out boilerplate when it makes sense.

Isn’t that what most developers do? Few are really inventing something new, afaict almost everyone is plugging together existing solutions to solve the same old use cases

Again; made no comment about what people do day to day. Specifically commenting on the authors praise of the LLM doing "novel" work.

Reply on news.ycombinator.com