The "next token" thing is literally true, but it might turn out to be a red herring, because emergence is a real phenomenon. Like how with enough NAND-gates daisy-chained together you can build any logic function you like.
Gradually, as these LLM next-token predictors are set up recursively, constructively, dynamically, and with the right inputs and feedback loops, the limitations of the fundamental building blocks become less important. Might take a long time, though.
Bill Gates was famous for this, about thirty years ago; Joel Spolsky wrote about it in https://www.joelonsoftware.com/2006/06/16/my-first-billg-rev.... Maybe it was over the top, and I'm sure it didn't work all the time, but I feel like it would have contributed to Microsoft's success.
Gradually, as these LLM next-token predictors are set up recursively, constructively, dynamically, and with the right inputs and feedback loops, the limitations of the fundamental building blocks become less important. Might take a long time, though.