Hacker News Reader

My perspective is a bit different. I've been fiddling with image generators a fair bit and got pretty good at getting particular models to generate consistently good images. The problem is a new model comes out every few months and it comes with its own set of rules to get good outputs.

LLMs are no different. One week ChatGPT is the best, next is Gemini. Each new version requires tweaks to get the most out of it. Sure, some of that skill/knowledge will carry forward into the future but I'd rather wait a bit for things to stabilize.

Once someone else demonstrates a net positive return on investment, maybe I'll jump back in. You just said it might take a year to see a return. I'll read your blog post about it when you succeed. You'll have a running head start on me, but will I be perpetually a year behind you? I don't think so.

Is that backed by any evidence? Taking 2 hours to perform a 1 hour task makes you half as productive. You are exchanging that for the uncertain prospect that it will be worth it in the long run. I think it's more likely that if you take 30 minutes to do the task in one year from now it's because AI got better, not because you made the code more AI friendly. In that case, those people taking 1 hour to perform the task now will also take 30 minutes to perform them in the future.

>This process will probably lead to a situation where in a year, it will take me 30 minutes with AI to do a task that would have taken me an hour otherwise

How do you figure?

>Because during those 2 hours I will make my code base more understandable to help the AI cope with it.

Are you working in a team?

If yes - I can't really imagine how does this work.

Does this mean that your teammates occasionally wake up to a 50+ changes PR\MR that was born as a result of your desire to "possibly" load off some of the work to a text generator?

I'm curious here.

> How do you figure?

Extrapolation. I see the progress I already made over the last years.

For small tasks where I can anticipate that AI will handle it well, I am already multiple times more efficient with AI than without.

The hard thing to tackle these days is larger, more architectural tasks. And there I also see progress.

Humans also benefit from a better codebase that is easier to understand. Just like AI. So the changes I make in this regard are universally good.

In my experience, only entry-level programmers are inefficient enough that using AI would double their productivity.

At the senior level or above, AI is at best a wash in terms of productivity, because at higher levels you spend more of your time engineering (i.e., thinking up the proper way to code something robust/efficient) than coding.

How long have you been doing this and how much has your time spent doing with AI what you could've done alone in an hour reduced as a result?

Or maybe they just tried it thoroughly and realised generative AI is mostly smoke and mirrors and has very little to offer for substantive tasks.

I hope your doubling of productivity goes well for you, I'll believe it when I see it happen.