My assumption is that every LLM-written article would just not have been written otherwise. That is a net good. It's also true that it's grating to read LLM-written prose.

Am I utterly mistaken or was LLM-written prose more tolerable when it was GPT-4o and Opus 3.x writing it?