it's not just the usual arguments about generative machine models and their social and economic and environmental costs. it's that the reason they are even superficially useful to someone like Terry Tao today is because they are plagiarising a huge wealth of explanatory text that mathematicians (including Tao himself, prolifically!) have spent generations amassing and sharing, and transitioning from the collective project of making and sharing these texts to a model of individuals consulting language models is making a very dangerous menu out of the discipline's seedcorn.