There was a nice thread on why Yann has put forward his theory of Autonomous AI. Somebody commented this is infact nice that now we have something concrete to work with, either to add, modify, delete elements from the theory. This is similar to Newton's method of approximation in numerical analysis. You start with initial guess (initial research paper) and then modify the original guess based on what worked and what didn't. Descartes got his original theory so wrong but Newton corrected it.