Hacker News Reader

Comment by epolanski | Hacker News Reader

I'm always surprised about the comments in these threads.

Genuinely seems many feel threatened by these tools that they need to emphasize the shortcomings.

And don't forget some random line about having studied algos in college and being able to reverse a tree, ugh...

Imo there's a difference between criticizing industry use and learning. If you haven't learned to code, but just to tell an LLM what you want, you will not be a successful software engineer.

There's no difference here in my mind between cheating and the models. If someone holds your hand for all your homework in school, you're not going to learn, whether that's a person or a language model.

So if AI advances in 4 years to the point where we're all deprecated, then fine, a student learning it got lucky. But otherwise, you're setting yourself up for failure if you lean too heavily on LLMs.

Example: Using a calculator in school when learning how to add/subtract/multiply/divide.

As a daily user of ai coding tools, I think if you don’t have criticisms of the tools you probably aren’t using them on production code.

I don’t think many would describe Yann Lecun as threatened by these tools, here’s his take on the LLM approach: https://youtu.be/MiqLoAZFRSE?si=415gg2vhjz2Sd84s

I am not arguing they have plenty of shortcomings rather than the very defensive attitude of many devs around here.

I'll speculate bluntly on the reason: the overemphasis many have put on the interview skillse is the reason why many feel threatened in the first place. Because that "skillset" is among the most replaceable with AI advances.

Speaking as someone who is defensive about it, it’s because most people pontificating about the AI takeover of programming aren’t actually in the industry or they’re the ones making the tools and have a vested interest in it actually happening.

IDK, it seems to me that the people who genuinely feel threatened by AI, are the ones that are impressed by it and/or fear it will only get better. Whereas those pointing out the shortcomings, and/or know it will always have its shortcomings, don't see any reason to feel threatened.

I really wonder about the programmers who feel threatened by LLMs. The vast majority of my daily work in software engineering does not remotely boil down to "write code for a discrete isolated task". It's all about solving difficult problems in a complex, interconnected system. Feature requests which require previous assumptions to be thrown out and new architectural decisions to be made.

Can current LLMs help with that? Almost never. What about in 5-10 years? Probably a little more, but I'm extremely skeptical that they'll ever reach beyond the tip of the iceberg.

I can only think it stems from fear of the unknown, leading people to believe in the worst-case scenarios.

But the way I look at it is: no matter how good AI gets, for AI to do something, someone must train it, instruct it, or ask it to do something for them, and then verify the results. All of which takes work from people. I mean even the mere instruction of saying "hey AI, do my taxes for me" takes work from me as I had to muster the very instruction, and there's no getting around this. So there will always be software developers regardless of however good AI gets. We may stop calling such people software developers, but I don't see why that matters.

Now, as whether we need less software developers because of efficiency gains? Unlikely to happen, because jobs are created from demand, and humans capacity for demanding things is endless, so the capacity for jobs is endless. This is the reason why the Jevons paradox happens. But in the case that we do in fact need less software developers, it still doesn't matter, because all it does is free up the time to do another job, just like a dishwasher does...

I'm not in big tech and I am absolutely worried because middle management and the c suite think my job will be redundant soon, even though I can see it won't, and thus they'll lay me or my peers off before shtf and ofc they fave no consequences for that decision.

May be tech managers are better, but seeing the recent run of layoffs, it seems they're not really that much better. The problem fundamentally is your boss doesn't know how to code and will be so impressed that he can get code to write fizzbuzz, so why does he need you anymore?

I don't feel threatened as a senior software developer. Given the kind of output I've seen from the current generation of these tools, their thoughtless use at scale will, if anything, create more demand for people who have enough understanding of how things work to clean up the inevitable mess that they will produce.

I am, however, extremely concerned as a user, because the senior management in many places will push for them anyway, with the resulting drop in software quality across the board.

I'm not threatened, juniors are because they won't learn. Also I do spend hours reviewing their fake code and trying to list all the bugs inside it instead of blindly accepting it. It's not my job to do this, it's dangerous and a waste of time.

Also I do use algorithms every day.

> Genuinely seems many feel threatened by these tools that they need to emphasize the shortcomings.

I think you're projecting a little bit.

You're not skilled or experienced enough in the field to understand that programming sometimes goes beyond simply gluing together snippets of simple, generalized code until you have something that sorta works, and thus believe AI to be more than capable of replacing you, and you don't understand why other, more experienced people don't feel the same way.

Reply on news.ycombinator.com