Remix.run Logo
▲ bayarearefugee 32 minutes ago

> There is no shortage of problems that can be solved (in fact, wants are infinite), and LLMs may actually enable our profession to reliably deliver them better than 50% completion.

Seems like copium to me if we're being honest.

Why would anyone pay you to solve the problem when they can just throw it at a cheap LLM?

And if the answer is that you have some sort of special taste or domain knowledge, once your solution is out in the world how are you going to stop someone from using an LLM to copy it and then undercut you in an immediate race to the bottom for everything?

▲kenferry 5 minutes ago | parent | next [-]

I agree with poster regarding "we're not running out of problems, and therefore we're not running out of jobs". I'm not sure whether the jobs will be "programming" though. And if I had to bet, I would say they will not be. I would guess that domain experts will be able to handle their own programming tasks most of the time. Which is cool! The same way people don't need dedicated typists anymore to transcribe text.

My grandfather worked at NASA, and my impression is that a significant portion of what he did was actual calculation. Nasa still employs people, they just don't do that particular work anymore.

▲ccppurcell 24 minutes ago | parent | prev | next [-]

If that analysis is correct (I tend to agree with it but I'm not sure) then we're heading for a world where no company can exist except for anthropic, openai etc. because if you're solving some problem with LLMs, and I am your customer, I can solve the same problem with LLMs and cut out the middle man. At the very least, there will be a race to the bottom until nothing is left.

If that happens I guess the whole thing hinges on how powerful and user-friendly open source models become.

▲johndhi 7 minutes ago | parent | next [-]

other options you're forgetting:

if your scenario starts happening, we break up anthropic and openai for antitrust reasons, or people do other things so that they don't become kings.

▲psadri 19 minutes ago | parent | prev [-]

I think that is essentially the bet around Anthropic / OpenAI

▲californical 5 minutes ago | parent | next [-]

Which is weird because there are so many things in society that will break up these companies rather than allow them to become bigger than governments.

It’s not possible for them to grow as big as their boosters want them to. There would be so much resistance, and if that resistance fails, then we’d already be in the middle of techno-dystopia hell anyways, where money wouldn’t matter.

▲TyadorB 7 minutes ago | parent | prev [-]

Why would they not be replaced as well?

▲tarballing 20 minutes ago | parent | prev | next [-]

Because those people have their own jobs and they don't care enough about software to treat it like a second job

▲deadbabe 28 minutes ago | parent | prev | next [-]

You create secret workflows, skills, and solutions that you never give others access to. Already doing this at work, 65% of my tooling is not available to any other employee.

▲coffeebeqn a minute ago | parent | next [-]

One of our staff engineers tried that but I just took one of the outputs they shared and told the same model figure out how to prompt and what to set up to do this and his super power was mine in an hour or less

▲_aavaa_ 24 minutes ago | parent | prev | next [-]

As they say, this doesn't seem like a durable moat.

▲kodoman 25 minutes ago | parent | prev [-]

I agree but then you have to make sure your prompting of the model does not expose the special technique, if your running local fine otherwise someone can just work out your tools or simply just pinch you skill markdown files.

▲johndhi 8 minutes ago | parent | prev | next [-]

call me crazy but you appear to be suggesting:

(1) LLMs can solve any human problem; (2) that #1 would be a bad thing if it were actually true because some SWEs can't get the same job they had before

if the LLMs are that great, tell them to find the SWEs a new, better job. if they aren't, and LLM+SWE is > LLM, then #1 falls apart.

▲arrowleaf 26 minutes ago | parent | prev [-]

I can see that you've been using Claude too much lately if you only read the first sentence/paragraph to decide what the entire comment says. You must not be working on sophisticated enough problems to see the rip current form as described. It (technical and cognitive debt) exists, just below the surface, and once it takes hold there's no swimming back to shore (a stable maintainable system) without extreme effort (refactors and first principles redesigns).