OK, HERE IS MY BEST GUESS ON THE STATE OF LLMS

Summary from the TLDR newsletter

Making a jump similar to the 100x jump going from GPT-3 to GPT-4 is going to be very hard. It will require a lot more training data and larger data centers. It is difficult to see whether efforts to improve models will bear any fruit. Policies like SB 1047 threaten to further slow down progress.