Steelman · slot A
The monoculture-of-language case
A computational linguist studying LLM diffusion would argue —We now have measurements, not just intuitions. The USC analysis of scientific journals, local news, and social media shows stylistic diversity collapsing after ChatGPT's release, and the Max-Planck team found ChatGPT-favored words like 'delve,' 'meticulous,' and 'boast' surging across 740,000 hours of spoken content. This is a feedback loop: the model was trained on us, and now we're being trained on the model. Even people who never open ChatGPT absorb its cadence secondhand, because that cadence has become the ambient register of 'professional' prose. A language with fewer stylistic attractors is a language that thinks in fewer directions, and we are watching that compression happen in real time.