Hacker News Reader

Comment by fritzo | Hacker News Reader

We should be more diverse with the software we build.

Any prompt jockey can make a faster ripgrep. The spec is "faster ripgrep".

What is the one thing that only you can specify, where your competitive advantage is maximized? Find that thing and pursue it with all your energy!

View on HN

If you didnt before you won't start now. AI does not speed up this kind of effort enough - its interpolating between existing stuff.

I disagree. There are things which are very annoying outside of core ideas or otherwise take a lot of time. Too much time and energy to justify normally. Personal example: https://github.com/wiwa/gpu-typing

Typechecking with my own algorithm. Simple enough to specify in Python. Annoying with much boilerplate in CUDA. Similarly annoying to test.

the threshold to start is definitely lowered now. So i disagree strongly.

the difficulty now is finishing.

However you can set up loops where the AI do improving by themselves and you just have to steer

> the difficulty now is finishing

Sigh. It was always like this.

From personal experience, yes you actually will. Speeding up small pieces of a design that's been in your head perhaps literally for years, is an incredible motivator. So is having a "pair programmer" to bounce ideas off of, 24/7 on demand.

I agree with the sibling comment: the difficulty [now, but also before] is finishing.

This past year, with agents' assistance, I've implemented a lot of ideas I had since 2022 or so. The quality isn't exactly where I had hoped but I improve that part over time and it feels nice to finally have something down in text files.

But this is precisely it - when you combine some kind of fidelity with software development (so you can guide the LLM well), with some kind of depth of knowledge in a domain (probably a niche domain) you can do amazing things.

I'm currently experimenting with some custom hardware + software, composed in a way that seems novel, for a niche domain. This is with a new kid coming soon. There's just no way I would have crossed the activation threshold to do that without having LLMs provide assistance.

> What is the one thing that only you can specify, where your competitive advantage is maximized?

Can you give some examples where people have done that?

That sounds intimidating. When I started my coding career, I followed my passion. I wanted to advance free software. I wanted it to be less buggy and more user friendly. I had no advantage over an average person in the field, I didn't even have a computer science degree.

If somebody's passion is faster command line tools, let them make a faster ripgrep. First results may be disappointing, but those who don't give up will figure out how to do it right.

I think what the OP is getting at is that if you can only prompt the model and aren't bringing anything to the table yourself, you're going to get a generic answer that looks like all of the others. The point is to figure out what differentiates you as a developer. Yes, that will favor those with more experience, but that's always been the case.

The point is, if you have no more detail in your idea than "faster ripgrep" then you don't contribute anything meaningful. Any number of other people can come up with "faster ripgrep", and the LLM won't respond differently to that prompt just because you're the one issuing it.

> What is the one thing that only you can specify, where your competitive advantage is maximized? Find that thing and pursue it with all your energy!

Or just look around for people who are doing this and then tell the LLM "faster <thing that the other guy did>"

The quality drop off when vibe coding something novel is pretty sharp. LLMs lean heavily on the existence of relevant training material.

That occurred to me when the author of the article was impressed with how well GPT-6 could reproduce zstd. Like hiw many of the trillions of parameters are just pieces of zstd.

Perhaps if you only know some qualities of the thing you want, and don't have an idea of its "shape" in code. Novel projects are, at an implementation level, mostly going to be novel combinations of relatively boring techniques and processes.

Reply on news.ycombinator.com