Imagine we would actually manage to truly automate programming… like some magic language that is predictable and has a super high-level interface. You can zoom arbitrarily deep into the low-level details, combine things freely without boundaries, etc. Our work would mostly be done and I think lots of us would still be sad and anxious about it. I like programming, because it’s a synthesis of striving for excellence and dealing with absolute chaos… and I need that chaos a bit like Batman needs the Joker. Something would be missing without it. But I’d still work towards that goal regardless!
The other thing is that “work” is much more than the monetary value to a business: it’s purpose, socializing, challenge, fun, etc. In our capitalist society, these things are merely side effects and may be cut off at any moment if the monetary value is declining. Productivity is the ultimate goal and the capitalist assumes everything follows from that: prosperity, purpose, etc. I think that assumption is now falling apart, even economically: if millions of white collar jobs are gone, so is the purchasing power. Who needs those AI products then? Was productivity really the bottleneck? What is the purpose of our economy? As business leaders and CEOs are starting to celebrate the approaching death of our craft and complaining with entitlement that we’re not joining those celebrations… I’m amazed and shocked by how far apart our values have always been.
LLMs (the way they’re implemented, aligned and governed) have the property that:
they do not actually bring us closer to the “automate programming” vision
they are an extension of the inhumane side of capitalism and are accelerating the decay of our values (social and engineering wise)
I’m not questioning that the technology itself may have value, but that doesn’t really matter. What matters is how it transforms the world. And it doesn’t look pretty.
This is a cynical take that comes to mind from time to time but modern software engineering is at the bedrock of the predatory advertising economy, the erosion of third places, and automation of other people’s jobs. Plus, maybe not at the scale of LLMs, it’s propped by a hardware industry that spent decades stoking war in resource rich African countries to get cheap coltan. Moreover, most people make CRUD apps so we aren’t really changing people’s lives with our precise engineering compared to, say, civil engineers. There’s a strong argument to be made for how. So maybe it’s accelerating a march that will refunnel some of the world’s brightest minds from figuring out how to make kids click on ads back into things that matter.
I don’t even understand why people are so up in arms about this LLM stuff. Nobody can push anything on you if you do not agree to it.
Since I’ve learned Haskell, I’ve mostly been working my own gigs where I could choose the tech I’m working with, simply because I like programming in Haskell. That freedom comes with a certain cost of course, something a lot of people aren’t prepared to pay.
What I’m seeing recently can only be described as a bunch of people who don’t want to pay the cost of freedom, trying to force people who own or manage companies to align with their goals and comfort zone.
If your boss says you have to use LLMs, you are free to quit if you can’t convince him otherwise.
Find another company that aligns with your ideals or open your own shop and show them how it’s done.
When the coding job is gone from the industry there will be recreational coding and coding as a performative art. I may be opening a coding gym even. I’ll name it “Fitness Function”
I don’t think this is an accurate representation of the situation.
“Pushing” doesn’t have to happen at literal gunpoint.
Nobody is forcing me to own a car; but with the state of public transit infrastructure, the demands of a typical 9-to-5 job, the legal situation around unemployment and the requirement to accept jobs that involve a serious commute, the way our cities are laid out, etc., not owning a car often implies some serious disadvantages. It’s not a hard “do this or I will hurt you” push, but there definitely is a push.
Likewise with AI coding. You are free to quit over your boss saying you have to use LLMs, but most of us have finite savings, and can’t just quit their jobs on a whim without repercussions. Some of us legit cannot afford the cost of freedom.
No. You just don’t want to pay the price of freedom. The “cannot afford” carries a lot implied meaning. What this usually translates to is “I don’t want to loose conveniences I’m used to”.
Also, who says you have to quit on a whim? Tighten up your belt a bit, save up, look for opportunities while working on your current job, when you are reasonably ready, inform your boss in a timely manner as a professional should, state your reason for leaving and leave.
Unfortunately this kind of dispute cannot be solved via rational argument. There are two competing notions of freedom that it is probably impossible to reconcile. @mastarija is thinking of negative freedom and @tdammers of positive freedom.
Yes, that’s probably right. Also, a small example of why LLMs can’t really be a good solution for every programming problem. Even humans don’t agree about what words mean exactly.
Actually LLMs are superhuman at working with this fuzzy concepts domain. Where the humans need an easy way out into “simple terms” and “definitions” the models can just superposition everything and bask in it. Humans can do that too, but not on the system-2 level.
That’s not really my experience. When I’m playing with it, and bouncing some ideas off of it, it very often completely misses the point and meaning behind my words. Perhaps I’m really bad at expressing myself, but I somehow don’t think that’s likely.
Which models did you use? Smaller models can have an embedding problem, but the larger ones are generally fine. Then there’s of course their RL training that interferes with the meanings. So shitty benchmaxxing can turn an otherwise good model into a abbrasive whacko.
