It's easy. Take for example the copper market (~$300B/year) and multiply at least by 10. Basically, it'll go everywhere, and more with the new applications. The only limit will be production.
Most places I've been are before Google maps, or even Google existed. I don't think timelines supports historic visits, and is unlikely to be simple to do.
There are some confusion. Competitive programming is not about optimizing CPU time, but about optimizing programmer time:
Competitive programming is a type of programming competition where participants compete against each other to solve algorithmic and computational problems in a limited amount of time (usually a few hours). The competition typically involves a set of problems of varying difficulty levels that participants must solve within a fixed time frame. The objective is to solve as many problems as possible and score the maximum number of points.
Competitive programming is often used as a way to test and hone programming and problem-solving skills, and it is popular among students and professionals in the computer science and software engineering fields. It can also be a fun and challenging way to learn new programming concepts and techniques.
While it may look at first a good idea for an enterprise to have such programmers that are able to program fast, I'd be wary about the code quality, maintainability, or even merely its adequacy to user specifications. And yes, about run-time too.
One can argue that if a competitive programmer finds a solution quickly, he has time left on his hands to fulfill those additional goals. It's indeed the argument used eg. by Lisp programmers, where the speed of reaching a solution is helped by the quality and high-levelness of the programming language, rather than brute competitiveness of the programmer. Then time is left to optimize the program if there are parts, or algorithms that are too slow for the user. But remember than the best gain of run-time performance are obtained by choosing the right data-structure and algorithms, rather than micro-optimizing and hacking.
There are multiple dimensions and notions behind the word "performance". Recently, energy efficiency has become an important performance parameter (you may choose an algorithm because executing it consumes less energy). Of course, companies may need programmers competent in various performance optimizations. Among other competencies...
Well, that's the point: you have to find a subject that doesn't bore the kid to death, and show him how he can explore it with programming. Any kind of programming. So what interests this kid?
You may also re-license under AGPL all your old software that you distributed so far under another license, as long as you still have the copyright on it. You won't be able to prevent others to use the old license, but you are in no obligation to continue distributing your old software under those old licenses. Of course, people who want up to date up-streams will have to agree to the new licensing terms. They may still fork the old software and keep it under the old license, but if they try to merge an AGPL patch, it'll become AGPL automatically by the GPL contamination. So they're quite incentivised to switch to the new upstream with the new AGPL license.
But it does. There's a famous experiment, where we display on a screen successive dots (with gaps in between). People see one dot in motion. Now, if half of the dots are red, and the following half are green, and if we ask people to show where the dot change color from red to green, they show the position right in the middle between the last red dot, and the first green dot. This goes to show that brain fills in missing frames, not only in space, but also in time, inventing a time in the past where the color of the dot changed.
This argument is correct, LLM are not AGI. However, that doesn't remove any utility from LLMs and ChatGPT. Only you should keep humans in the loop. It's a "chat" architecture, ie. you must have a dialog, between a human and the LLM. (or possibly some other AI components and the LLM).
While ChatGPT alone couldn't re-invent the Velcro or something similar, you could still use to brainstorm and come with some innovative solution to your problem. See for example, how such a conversation might go:
