I think the fallacy of LLM code assistants is that it feels as if you're going fast and they're doing a lot of work. But then you have to validate them, correct their course, and you end up with a ton more work, still typing it all out anyways. It do be addictive though. - ThreadSky

About ThreadSky

pvdz.ee • 6 days ago

I think the fallacy of LLM code assistants is that it feels as if you're going fast and they're doing a lot of work. But then you have to validate them, correct their course, and you end up with a ton more work, still typing it all out anyways.

It do be addictive though.

Comments

creationix.blue•6 days ago

I finally found one place where they are great; writing node.js native add-ons for common C libraries.

Sure it takes a ton of time back and forth fixing errors, but most of the better models are good enough to figure it out eventually.

But most of my hobbies are too esoteric for them to help.

creationix.blue•6 days ago

The result is never production quality, but it does go from zero to semi-functional proof of concept very quickly. That's where I typically got stuck in the past. So much initial friction with native add-ons.

pvdz.ee•6 days ago

True, but it's hard to realize the right point of hopping off and just completing the work yourself.

creationix.blue•6 days ago

I kinda want a halfway integrated agent that can create and update files directly and see my terminal commands and the resulting errors, but I still manually run the program

pvdz.ee•6 days ago

That's what cursor basically does.

I want a sandbox where I can let the agent run a particular command, generate test files, and fix the code until all tests are passing etc.

This should be great for a parser-sort of library where tests are simple input/output.

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply