In general, I think you make your money in Tech at the ends.. Either bleeding edge or trailing edge. ATM, I'm working on Rust, serverless, playing with Pijul, keeping an eye on OpenTF, and pulumi.
For trailing edge stuff - I figure I'll be the worlds oldest COBOL coder by the time I retire and can use that to fund my retirement :-P
(And yes, COBOL will be with us for many years to come. In the next century, we'll probably be emulating IBM mainframes on quantum computers.)
including current versions quality degradation, unprofitable business models (it's still too damn expensive to be covered by your average 19$/m subscription) etc.
It's a common phenomena, described with an S-curve: (https://en.wikipedia.org/wiki/Sigmoid_function). We can't know if LLMs are already there, or not yet, we'll only know after the fact.
At best it would embrace using some AI assisted tools in no more extent than today's cars use algorithms for those safety features.
But the nature of the profession will only change when the mainstream computers design changes drastically (and there is no strong signs of that yet)
If ever, a GPT-like LLM/AGI ever exists where it can distill business requirements, understand modular designs and intelligently establish complex relationships between different systems and contexts, then 99% of all jobs will perish.
This will be an unprecedented disruption at a macro scale that humanity has never seen before. All of our current economic models will be instantly trash. How likely do you think this will happen? If it does, there's no real incentive to have these systems produce anything because no one can afford to buy anything. A global revolution would be inevitable.
Exactly.
How likely do you think this will happen?
I estimate the chance of this happening after the release of GPT-5: 50%. Chance of this happening after GPT-6 is released: 90%.
You painted a pretty bleak picture of what might happen. I tend to agree with your assessment, but there's a chance that the total automation of everything will make things so cheap that quality of life will actually improve for majority of humankind. In any case, we still have a few years to prepare for this disruption.
p.s. I'm not sure why people are talking about software engineering as some sort of a subtle skill that is hard to automate. It's quite likely that GPT-5 will be able to talk to my boss, colleagues, and any other other stakeholders to understand what needs to be done, access all relevant information, iterate on requirements, ask for help, and problem solve, and it will do so better than me, or anyone else in this thread. Even if GPT-5 is not quite there yet, it's almost guaranteed that GPT-6 will be (barring some unlikely global catastrophic events preventing further technological progress). At that point, there will be no reason for my boss to employ me, or for his boss to employ him, and so on.
The LLMs that we have today are amazing, but there is still plenty of room for improvement. Having to train it on a huge dataset is problematic for some uses; perhaps there is a related structure that can be trained more easily and more quickly. That would also reduce the effect of OpenAI's monopoly. LLMs also have specific weaknesses, like poor performance at arithmetic. At this stage, I wouldn't really feel comfortable feeding problems into an LLM and presenting users with the LLM's answers. It's still the Wild West in many aspects. There is always an improvement on the horizon, but it's hard to tell where it will come from and when it will come. Maybe we'll have LLMs that really start to resemble intelligence, or maybe we'll have a totally different structure that does everything LLMs can do plus more.
By the time most of today’s wannabe AI gurus are good enough to get a decent job offer - AI would not be such a big deal (best case scenario imo).
I didn’t notice any “obvious” changes in the game that weren’t there before.
For instance I don’t see anything different in requirements and comments of HRs reaching me.
JVM is still same jvm, bugs are still bugs, unit tests are still unit tests.
The fact that now instead of googling a problem you google and gpt it is a minor change in any real world complex engineering context that im aware of.
What exactly is obvious to you that im missing?