Remix.run Logo
▲ deltarholamda 6 hours ago

People now may be able to understand it. I'm very concerned about the future programmers who are 10, 15 years old now. If you have never built by hand, how are you going to understand the code emitted from an LLM, or whatever New Thing we come up with in 10 years?

Back in the 90s, places that hired 3D modelers/animators would say something like "Look, we don't care if you know the program. We can teach you that. What we want to know is can you draw?" There's a base level understanding required for any higher-level endeavor. I would not expect a person who might even be considered an expert Python programmer to immediately be ready to become a core maintainer for the kernel. If you don't do the low-level stuff, early and often, you won't recognize the foot-gun when you see it.

Maybe the LLMs will get so good that isn't a problem? Okay, sure, possibly. If it gets to that point before all the greybeards retire or die, then it's possible we can dodge what I see is a negative feedback loop. If not, then I think the future is much less rosy.

▲tl 5 hours ago | parent | next [-]

In the 90s, we had a sane split between people who coded the tools those animators used in C++ and everyone who got leverage out of BASIC or SQL or Excel. These two groups self-sorted by passion or access to a full computer science education. A majority of software writers have been building mediocre line-of-business software for decades leveraging a poor understanding of code.

LLMs emitting code are a great workaround for "bury the complexity, I just need the result" and terrible for "this codebase produces long-term value". People are slow to learn the difference, and the AI gold rush profits from the confusion. We thought television would bring Shakespeare to the masses. It didn't, but theater was still there for Patrick Stewart (RIP). Software engineering will be okay.

▲dam_jackalopes 3 hours ago | parent [-]

I'm still in two minds about this. I'm firmly in the "leverage it" group because I'm here to do my job, make the org money and go home. Agentic coding is great for this though it will likely thin the herd, so to speak, at least in the short term.

I'm not at all convinced that LLMs will kill software engineering but I'm equally unconvinced that it will be an overall positive for innovation in the field, at least while we're basically all running things through the same handful of closed model orgs. It's just a really weird time where I can't decide if I'd rather jump back and work in the world of five years ago or take my chances and jump forward the same period.

▲popkkksidhd 5 hours ago | parent | prev | next [-]

Even with developing, I’ve noticed understanding “beneath” where you’re working greatly accelerates LLM assisted and non LLM assisted work. I think that’s the true way to get good at using LLMs - any other “prompt engineering” kinda stuff is snake oil.

And yeah, I’m not sure how that goes away either - life is full of tradeoffs, how do you RL things with no objective truth?

▲FeteCommuniste 4 hours ago | parent | next [-]

The paradox of LLM usage: the best way to use an LLM well in a domain is to know the domain, but if you spend all your time using LLMs, your domain knowledge may atrophy.

▲lukeschlather 3 hours ago | parent [-]

The thing is you pick up different domain knowledge. I was using AI to experiment with computer vision stuff last week, and I learned so much about hardware sensors, about accelerometers, magnetometers. I learned about the relationships between pixels and millimeters when you're talking about cameras in some reference frame. I learned about how the speed of light relates to the time it takes for wifi and bluetooth to send a message through household objects. I learned about how sound reflects in spaces and how different frequencies are picked up by microphones in different kinds of hardware I have.

There's so much to learn in the world. The domain knowledge I used to carry about software, I hope I forget most of it. I hope I learn things that let me do things I never could've conceived of before.

▲FeteCommuniste 3 hours ago | parent [-]

The knowledge you gain by reading someone else's solutions is real but not as sticky as what you pick up by working out a solution yourself. I fear we are well on our way to the second type being a thing of the past.

▲lukeschlather 3 hours ago | parent [-]

I wasn't just "reading solutions" I was wandering around my house with different devices, repositioning them, taking video, thinking about how to figure out what was going on. The code is never the point, it's the means to an end, the problems I'm trying to solve, the code is only one piece among many.

▲deltarholamda 5 hours ago | parent | prev [-]

Understanding "beneath" is a good idea. I used the free version of Grok to generate some Ansible scripts. I don't use Ansible often enough to be really adept, and I appreciated some help.

It was pretty good, understood what I wanted to do, and broke it up in playbooks/roles/etc. quite well. But then it did things like continuously overwrite the LetsEncrypt cert for each domain with the next one in a loop. It understood "this needs to be a loop" and how to optimize that loop, but not that it was going to overwrite each domain's key with the next.

I'm a pretty good sysadmin, so I recognized this and could take what it did and fix it, but it required a deeper understanding of the underside of nginx/LetsEncrypt and the filesystem to even see what it was doing.

Of course, once you run the playbook, test a domain and see its SSL cert is all jacked up you'll go looking for WTF happened, but just blindly taking the output as given meant it was horribly broken.

It may be that Ansible, by its nature, is a bit sui generis because it touches so many unconnected domains, and therefore is harder than outputting Python. I don't know. But one needs more than a surface level understanding of those domains if you want to catch the LLM being stupid.

▲bluGill 4 hours ago | parent | prev [-]

The same thing was said when CAD and printers replaced manual drawing boards.

▲bemtevi 3 hours ago | parent [-]

This is nothing like that situation.