The jobs at risk / already dead from AI are things like internet ghost writing / SEO spam writing and customer service roles for cheap companies. For the rest of us it's a (very useful) tool, not a competitor.
But I don't think that's the real risk. The real risk is people in charge of jobs that believe AI can replace their employees, regardless of its true capabilities.
LLMs cannot even solve math problems, and we expect them to do tasks far more complex e.g. software engineering.
Attempts to make them good at math have shown its not easy, and often times sacrifices language quality. How do we expect something that cannot solve y=mx+b reliably for any input >100 to write more than a todo list app?
A lot of companies/investors are just assuming its a training data problem, but we can already produce massive amounts of robust math training data, and LLMs still aren't good at math.
At least this is my understanding, I was doing some research into Kahnmigo and its math tutoring capabilities and went down a rabbit hole about LLM math.
Who expects that? Even if LLMs produced perfect output every time, they are still just a fancy compiler. You continue to need a software engineer to write the code.
Some expect that they will enable software to go the way of the elevator, which is to say that modern elevators didn't eliminate the elevator operator – they made everyone the elevator operator. This is quite possible, and has arguably already happened at least in some niche cases. But that is an expansion of the pool of human software engineers, not an LLM doing software engineering.
Maybe there is some kind of technology out there that will someday take on the tasks of software engineering, but not LLMs. They cannot do software engineering any more than a C++ compiler can do software engineering.
https://www.cognition-labs.com/introducing-devin
SA: “Coding is probably the single area from a productivity gain we’re most excited about today"
Nvidia CEO: "Don't learn to code"
The whole group of companies working on it are all pushing that it will make programming redundant, but they don't even make a Ti-83 redundant.
Even if its work that a wider range of people could do, it's still work that's going to need to be done, and it's not going to be done by a chatbot. So it seems that even for many jobs that LLMs could ostensibly replace, it's more likely to widen the labor pool rather than literally replace people. If anything it might increase the number of businesses and opportunities, because it would enable the breadth of domains that micro and small companies could perform in.
Vehicles were quite unique in that they were capable of entirely replacing their predecessor for simple transportation tasks. But in e.g. software, the programming tasks LLMs may one day be able to carry out are but one aspect of the job. Getting back to janitors for instance, you're not going to replace a janitor with a roomba.