I knew it was over when we used automatic place and route to synthesize entire CPUs in Cadence in my ECE classes in 1999 before I graduated with my degree. It already achieved 80% of the circuit density and speed as humans. Then read Genetic Programming III by John Koza 5 years later, which explained that each computer 10 years older than the previous generation, running about 100 times faster due to Moore's Law, could be made to write previous gen software automatically.
Unfortunately, GPUs don't help much with the kinds of optimization problems that I was interested in, where we really need many thousands or even millions of independent threads exploring enormous search spaces using multicore CPUs with local memories. Gradient descent is one way to do that, but IMHO it's one of the hardest ways, requiring the most bare hands effort and brute force. That's why neural nets that actually worked took 25 years to manifest.
So imagine knowing how to basically grow computer hardware and software organically with self-training/improvement in the late 90s and early 2000s, only to be stifled by the end of Moore's Law arriving in 2007 when smartphones chased the profit in cost/power reduction of CPUs and devoted most all remaining R&D to faster SIMD GPUs which didn't help MIMD (multiple instruction multiple data) workloads.
After working so hard for so many years and decades, only to tread water making rent, some part of me welcomes watching the world burn. Now everyone else will get to enjoy the precarious feeling of having the skills to do real work, only to be forced not to at some BS job that almost pays the bills.
Unless you all prove me wrong and put your potential to real use, and bring time/money/resources to the working poor, outside of the current wealth inequality pyramid scheme we pretend to call capitalism.
Maybe I sound bitter, but I'm feeling oddly inspired and optimistic for the first time in a quarter century.