Peter Thiel gets a lot right, but boy did he nail it when he said the wordcels would actually be better off than the math and science bros due to AI. The underlying reason for this is the structure of language reveals and amplifies human preferences, whereas science and maths are verifiable domains searching for the truth. It really doesn't matter what GPT-10 can do, I will still shitpost because I like to post, and if you enjoy my worldview you will still enjoy reading it. Sure, GPT-10 will be capable of superpersuasion and genuinely insane insights, but it will not have the weights of my own neural network, so it doesn't matter. It's the same reason that I don't like asking an AI to summarise anything... To summarise something you need to know what my personal mental model of the world would think of the text/video being consumed. And since uploaded intelligence doesn't exist yet you can only do a very crude approximation of that. Even when continual learning comes online it will still be but a blunt simulation of my mind rather than my mind itself. Worth remembering this. Anything that relates to human preferences will be amplified due to AI as people seek the compliment.