A few points on AI/ML :
- Works best for Image-Video-Audio processing.
- Doesn't work that well for natural language processing, or text processing.
- You get big improvements in the early stage. After a certain level, increasing processing power x1000 times gives only a quality increase of 40%. Work is stagnant, not way you can squeeze more off AI/ML. That's the end of how much AI can do. And most of all work now are at this stage. The edge of its limit. Without any way to progress ahead.
- You need too many low precision floating point numbers for neural connection. Memory being the hog. Which is why they are trying to come up with new floating point standards optimized for such calculations.
- Development is now stagnant, or moving slow. New work is expected to come from engineering and implementation side, not science.
The arms race is over. We now use the inventions.