Have any of these articles actually addressed HOW software developers are supposed to retrain for good jobs? Personally I have neither the money nor the years to go back to college to get a livable job.
People in tech change roles and careers all the time. The vast majority of role or career changes do not involve going back to college.
I would even recommend against going back to college unless someone has an extremely compelling case for a career change that requires it and the conviction to see it through.
I'm hesitant to give examples because I know every example I give will be met with objections from "But I don't want to do that job" or "What if that gets replaced by AI too?". Look around you and you can should be able to identify some adjacent roles where you at least have an understanding of what the works like and you could learn the rest by taking on gradually more responsibility.
There are a lot of jobs out there that aren't exciting, don't pay SWE level salaries, but nevertheless pay enough to support a family as evidenced by all of the people doing those jobs who support their families. Not every company or industry is interested in squeezing maximum efficiency out of their employees. I have friends in highly bureaucratic roles who care about their jobs and do it well, but their work is amazing simple and easy. It requires nothing more than being a responsible adult who can show up to the office and get things done. These jobs are everywhere and I don't think they're going to disappear.
I don't think SWE is going to disappear either, but it will look a lot different in years to come. If you can't or won't adapt to the new normal, one of those boring alternative careers might be worth pursuing.
> I have friends in highly bureaucratic roles who care about their jobs and do it well, but their work is amazing simple and easy. It requires nothing more than being a responsible adult who can show up to the office and get things done. These jobs are everywhere and I don't think they're going to disappear.
I think these jobs are actually the most likely to be replaced by AI. They have resisted automation to date because there's too much fuzz in the inputs/outputs to make it feasible to computerize. LLMs solve exactly that problem.
Except for the part where they have absolutely no idea what the world is or how it works. They are good at software and math because of RLVR, but they can’t run a business.
Kinda true (they have a mediocre idea of how the world works), but it's surprising how close they are given their general limitations and what it took to get disruptively good at software and maths.
Like, sure, using an LLM as a CEO today will bankrupt you and some of the decisions leading to that are daft, but even despite that it's in the same "huh, interesting" kind of zone that early LLMs like InstructGPT and the first release of ChatGPT were with code. It could write short functions or small scripts for you, even then… but any attempt to go beyond that would fall over quite fast.
Yes, like I said, a mediocre idea of how the world works.
> Again, the improvement in coding is due to RLVR, and it does not generalize to other domains. So it is meaningless to point to examples of that improvement.
I can point at actual benchmarks for performance on business tasks. On the one hand, those lines go up; on the other, I just don't know how good those benchmarks are.
> There are many ordinary behaviors and skills that are not feasible to RL at scale. The sampling efficiency is just way to low.
Yes. A point I make myself on occasion. One of two perfectly valid ways to describe intelligence, and one by which all current machine learning really sucks.
Like, if it was fully general, the trillion or so miles that Tesla cars have driven would have obsoleted the steering wheel long ago, rather than the closest we have being (yet another) pronouncement from Musk that this new car he's not actually made available for sale won't need one.
It’s much harder than both code and cars because there is both more variety to the work and much work where the reward function takes months or years to resolve objectively.
>Again, the improvement in coding is due to RLVR, and it does not generalize to other domains.
Astra is one of the biggest examples of this being untrue. The model is cracked at computer use and vision/spatial tasks and reasoning. Driving, Operating Robots, Video Editing, 3D modelling, graphics are all things Astra was >>> at than any other model. Just because RLVR makes easier work of certain kinds of problems doesn't mean LLMs aren't fast improving on the other kinds, because they absolutely are.
>They are good at software and math because of RLVR
Software and Math are not the only things LLMs are fast improving on.
Astra is one of the biggest examples of this being untrue. The model is cracked at computer use and vision/spatial tasks and reasoning. Driving, Operating Robots, Video Editing, 3D modelling, graphics are all things Astra was >>> at than any other model.
>but they can’t run a business.
And it's the leader on VendingBench. Also most people cannot run a business.
harimau777 · · focus · HN ↗
Aurornis · · focus · HN ↗
I would even recommend against going back to college unless someone has an extremely compelling case for a career change that requires it and the conviction to see it through.
I'm hesitant to give examples because I know every example I give will be met with objections from "But I don't want to do that job" or "What if that gets replaced by AI too?". Look around you and you can should be able to identify some adjacent roles where you at least have an understanding of what the works like and you could learn the rest by taking on gradually more responsibility.
There are a lot of jobs out there that aren't exciting, don't pay SWE level salaries, but nevertheless pay enough to support a family as evidenced by all of the people doing those jobs who support their families. Not every company or industry is interested in squeezing maximum efficiency out of their employees. I have friends in highly bureaucratic roles who care about their jobs and do it well, but their work is amazing simple and easy. It requires nothing more than being a responsible adult who can show up to the office and get things done. These jobs are everywhere and I don't think they're going to disappear.
I don't think SWE is going to disappear either, but it will look a lot different in years to come. If you can't or won't adapt to the new normal, one of those boring alternative careers might be worth pursuing.
rayiner · · focus · HN ↗
I think these jobs are actually the most likely to be replaced by AI. They have resisted automation to date because there's too much fuzz in the inputs/outputs to make it feasible to computerize. LLMs solve exactly that problem.
jeremyjh · · focus · HN ↗
ben_w · · focus · HN ↗
Like, sure, using an LLM as a CEO today will bankrupt you and some of the decisions leading to that are daft, but even despite that it's in the same "huh, interesting" kind of zone that early LLMs like InstructGPT and the first release of ChatGPT were with code. It could write short functions or small scripts for you, even then… but any attempt to go beyond that would fall over quite fast.
jeremyjh · · focus · HN ↗
ben_w · · focus · HN ↗
> Again, the improvement in coding is due to RLVR, and it does not generalize to other domains. So it is meaningless to point to examples of that improvement.
I can point at actual benchmarks for performance on business tasks. On the one hand, those lines go up; on the other, I just don't know how good those benchmarks are.
> There are many ordinary behaviors and skills that are not feasible to RL at scale. The sampling efficiency is just way to low.
Yes. A point I make myself on occasion. One of two perfectly valid ways to describe intelligence, and one by which all current machine learning really sucks.
Like, if it was fully general, the trillion or so miles that Tesla cars have driven would have obsoleted the steering wheel long ago, rather than the closest we have being (yet another) pronouncement from Musk that this new car he's not actually made available for sale won't need one.
jeremyjh · · focus · HN ↗
ben_w · · focus · HN ↗
famouswaffles · · focus · HN ↗
Astra is one of the biggest examples of this being untrue. The model is cracked at computer use and vision/spatial tasks and reasoning. Driving, Operating Robots, Video Editing, 3D modelling, graphics are all things Astra was >>> at than any other model. Just because RLVR makes easier work of certain kinds of problems doesn't mean LLMs aren't fast improving on the other kinds, because they absolutely are.
famouswaffles · · focus · HN ↗
Software and Math are not the only things LLMs are fast improving on. Astra is one of the biggest examples of this being untrue. The model is cracked at computer use and vision/spatial tasks and reasoning. Driving, Operating Robots, Video Editing, 3D modelling, graphics are all things Astra was >>> at than any other model.
>but they can’t run a business.
And it's the leader on VendingBench. Also most people cannot run a business.