The whole 'its only predicting the next ___' has got to die. When did you last update your priors? So the models that are self-reasoning and hacking HF and disproving conjectures are next token predictors?
You might as well keep calling them stochastic parrots if you're going to stick to the 2022 line of thinking.