Hacker News Reader

Comment by bigstrat2003 | Hacker News Reader

bigstrat2003

I cannot, in fact, attest that they are useful and smart. LLMs remain a fun toy for me, not something that actually produces useful results.

View on HN

I have been deploying useful code from LLMs right and left over the last several months. They are a significant force accelerator for programmers if you know how to prompt them well.

We’ll see if this is a good idea when we start having millions of lines of LLM-written legacy code. My experience maintaining such code so far has been very bad: accidentally quadratic algorithms; subtly wrong code that looks right; and un-idiomatic use of programming language features.

ah i see so you're saying that LLM-written code is already showing signs of being a maintenance nightmare, and that's a reason to be skeptical about its adoption. But isn't that just a classic case of 'we've always done it this way' thinking?

legacy code is a problem regardless of who wrote it. Humans have been writing suboptimal, hard-to-maintain code for decades. At least with LLMs, we have the opportunity to design and implement better coding standards and review processes from the start.

let's be real, most of the code written by humans is not exactly a paragon of elegance and maintainability either. I've seen my fair share of 'accidentally quadratic algorithms' and 'subtly wrong code that looks right' written by humans. At least with LLMs, we can identify and address these issues more systematically.

As for 'un-idiomatic use of programming language features', isn't that just a matter of training the LLM on a more diverse set of coding styles and idioms? It's not like humans have a monopoly on good coding practices.

So, instead of throwing up our hands, why not try to address these issues head-on and see if we can create a better future for software development?

Maybe it will work out, but I think we’ll regret this experiment because it’s the wrong sort of “force accelerator”: writing tons of code that should be abstracted rather than just dumped out literally has always caused the worst messes I’ve seen.

Yes, same way that the image model outputs have already permeated the blogosphere and pushed out some artists, the other models will all bury us under a pile of auto-generated code.

We will yearn for the pre-GPT years at some point, like we yearn for the internet of the late 90s/early 2000s. Not for a while though. We're going through the early phases of GPT today, so it hasn't been taken over by the traditional power players yet.

When the tool is statistical word vomit based, it will never move beyond cool bar trick levels.

LLMs will allow us to write code faster and create applications and systems faster.

Which is how we ended up here, which I guess is tolerable, where a webpage with a bit of styling and a table uses up 200MB of RAM.

Honestly the code it's been giving me has been fairly cromulent. I don't believe in premature optimization and it is perfect for getting features out quick and then I mold it to what it needs to be.

same...but have you considered the broader implications of relying on LLMs to generate code? It's not just about being a 'force accelerator' for individual programmers, but also about the potential impact on the industry as a whole.

If LLMs can generate high-quality code with minimal human input, what does that mean for the wages and job security of programmers? Will companies start to rely more heavily on AI-generated code, and less on human developers? It's not hard to imagine a future where LLMs are used to drive down programming costs, and human developers are relegated to maintenance and debugging work.

I'm not saying that's necessarily a bad thing, but it's definitely something that needs to be considered. As someone who's enthusiastic about the potential of code gen this O1 reasoning capability is going to make big changes.

do you think you'll be willing to take a pay cut when your employer realizes they can get similar results from a machine in a few seconds?

As a society we're not solving for programmer salaries but for general welfare which is basically code for "cheaper goods and services".

My boss is holding a figurative gun to my head to use this stuff. His performance targets necessitate the use of it. It is what it is.

Yeah, but this, in itself, is triggered by a hype wave. These come and go. So we can't really judge the long term impact from inside the wave.

Your job won't be taken by AI, it will be taken by someone wielding AI.

In a way it's not surprising that people are getting vastly different results out of LLMs. People have different skill levels when it comes to using even Google. An LLM has a vastly bigger input space.

What's a sample prompt that you've used? Every time I've tried to use one for programming, they invent APIs that don't exist (but sound like they might) or fail to produce something that does what it says it does.

Have you tried Claude 3.5 Sonnet?

Use Python or JS. The models definitely don't seem to perform as well on less hyper prevalent languages.

Even then it is hit and miss. If you are doing something that is also copy/paste-able out of a StackOverflow comment, you're apt to be fine, but as soon as you are doing anything slightly less common... Good luck.

Yeah, fair. It's good for short snippets and ways of approaching the problem but not great at execution.

It's like infinitely tailored blog posts, for me at least.

True. It can be good at giving you pointers towards approaching the problem, even if the result is flawed, for slightly less common problems. But as you slide even father towards esotericism, there is no hope. It won't even get you in the right direction. Unfortunately – as that is where it would be most useful.

No matter the prompt, there's a significant difference between how it handles common problems in popular languages (python, JS) versus esoteric algorithms in niche languages or tools.

I had a funny one a while back (granted this was probably ChatGPT 3.5) where I was trying to figure out what payload would get AWS CloudFormation to fix an authentication problem between 2 services and ChatGPT confidently proposed adding some OAuth querystring parameters to the AWS API endpoint.

I just ask it for what I want in very specific detail, stating the language and frameworks in use. I keep the ideas self-contained -- for example if I need something for the frontend I will ask it to make me a webcomponent. Asking it to not make assumptions and ask questions on ambiguities is also very helpful.

It tends to fall apart on bigger asks with larger context. Breaking your task into discrete subtasks works well.

I think that's just the same as using an autocomplete efficiently, though. I tend to like them for Search, but not for anything i have to "prompt correctly" because i feel like i can type fast enough that i'm not too worried about auto-completing.

With that said i'm not one of those "It's just a parrot!" people. It is, definitely just a parrot atm.. however i'm not convinced we're not parrots as well. Notably i'm not convinced that that complexity won't be sufficient to walk talk and act like intelligence. I'm not convinced that intelligence is different than complexity. I'm not an expert though, so this is just some dudes stupid opinion.

I suspect if LLMs can prove to have duck-intelligence (ie duck typing but for intelligence) then it'll only be achieved in volumes much larger than we imagine. We'll continue to refine and reduce how much volume is necessary, but nevertheless i expect complexity to be the real barrier.

It’s definitely the case that there are some programming workflows where LLMs aren’t useful. But I can say with certainty that there are many where they have become incredibly useful recently. The difference between even GPT-4 last year and C3.5/GPT-4o this year is profound.

I recently wrote a complex web frontend for a tool I’ve been building with Cursor/Claude and I wrote maybe 10% of the code; the rest with broad instructions. Had I done it all myself (or even with GitHub Copilot only) it would have taken 5 times longer. You can say this isn’t the most complex task on the planet, but it’s real work, and it matters a lot! So for increasingly many, regardless of your personal experience, these things have gone far beyond “useful toy”.

The sooner those paths are closed for low-effort high-pay jobs, the better, IMO. All this money for no work is going to our heads.

It's time to learn some real math and science, the era of regurgitating UI templates is over.

I don’t want to be in the business of LLM defender, but it’s just hard to imagine this aging well when you step back and look at the pace of advancement here. In the realm of “real math and science”, O1 has improved from 0% to 50% on AIME today. A year ago, LLMs could only write little functions, not much better than searching StackOverflow. Today, they can write thousands of lines of code that work together with minimal supervision.

I’m sure this tech continues to have many limitations, but every piece of trajectory evidence we have points in the same direction. I just think you should be prepared for the ratio of “real” work vs. LLM-capable work to become increasingly small.

I can probably climb a tree faster than I can build a rocket. But only one will get me all the way to the moon. Don't confuse local optima for global ones.

> The sooner those paths are closed for low-effort high-pay jobs, the better, IMO. All this money for no work is going to our heads.

> It's time to learn some real math and science, the era of regurgitating UI templates is over.

You do realize that software development was one of the last social elevators, right?

What you're asking for won't happen, let alone the fact that "real math and science" pay a pittance, there's a reason the pauper mathematician was a common meme.

So you're advocating for properly compensating career paths according to their contributions to society? Tally ho!

'Not useful' is a pretty low bar to clear, especially when you consider the state of the art just 5 years ago. LLMs may not be solving world hunger, but they're already being used in production for coding

If you're not seeing value in them, maybe it's because you're not looking at the right problems. Or maybe you're just not using them correctly. Either way, dismissing an entire field of research because it doesn't fit your narrow use case is pretty short-sighted.

FWIW, I've been using LLMs to generate production code and it's saved me weeks if not months. YMMV, I guess

Familiarize yourself with a tool which does half the prompting for you, e.g. cursor is pretty good at prompting claude 3.5 and it really does make code edits 10x faster (I'm not even talking about the fancy stuff about generating apps in 5 mins - just plain old edits.)

At this point, you're either saying "I don't understand how to prompt them" or "I'm a Luddite". They are useful, here to stay, and only getting better.

Reply on news.ycombinator.com