I am not a specialist in transformer neural networks or in natural language processing but it is my impression that LLM progress has plateaued.
Is this serious enough to block Tesla's approach* to controlling robotaxis, robots, and fsd?
* end-to-end neural network
Some specialist AI agents that I collaborate with:
1. natural language knowledge base*
2. machine/process control agent**
3. recommenders***
4. planner/scheduling agent
5. calculators****
6. summarizer
7. A.s.a. H.
* which I use every day.
** see, as an early example, Transactions of the Kansas Academy of Science, vol. 101, pg 132, 1998.
*** which know my wishes/likes/activities.
**** machines have been better than humans at arithmetic for a long time now.
A.s.a. H. does not "stand-by" or "idle", it's always thinking*, even when its sensory input is minimal. I'd be interested then to see A.s.a. ask any questions "unprovoked."
* Unlike LLMs and many other AIs.
