This article is largely wrong on the facts and history. I was there. I'll swear to it. Physicists should not be taking authoritative views on AI and computational cognitive science and linguistics any more that they should be taking authoritative views on physics. They are simply making the history, and present, up.
Here is an early advanced perceptron model I coded in 1980 which was fully parallel to match observable facts about eye movements in human reading behavior. Same computation as in LLMs. If you want to understand the truth ask the people who lived it. Sadly, Jaime Carbonell is dead now, as is Roger Schank. And Chomsky has been proven right although he never had a sense of computation. Others did.
https://link.springer.com/article/10.3758/BF03203564
I do concur with the idea that listening to people with a commercial interest in the game is not a good idea. When Paul is writing about Physics, his stuff is very good for us amateurs, at least, and I suspect real physicists as well.
Chomsky is right that we are still a long way from the AGI, but its not that we do not know how to do it. We have, for a long time. It is that the hardware for modelling still does not exist so the operational models are subject to existing computing hardware constraints. It has gotten a lot better than in 1980, but is still a long long way off from the actual computation that routinely goes on between your ears.
For example, we ran a railroad car recognition system based entirely on advanced perceptron theory for six months in 1991, and it never once failed to have perfect performance, with one trial learning on all 4 Million RR Cars in North America, with perfect real-time recognition (zero classification error, type 1 or 2) in six months of daily operation. When we applied similar pattern recognition on Chinese Character Recognition, we had the best in the world, in 1994 (30 years ago). These were niches because the hardware was not, and is not, available for fully general AGI.
My hardware systems for convolutional feature detection for advanced perceptron classifiers were used worldwide for scanning and reading newspapers and other types of material from Hardware I built in 1983.
Here is some authoritative stuff to read by someone who lived it (as a scientist/engineer).
.
https://medium.com/liecatcher/100-billion-sources-of-hate-between-your-ears-d34baf503c98
