The AI Content Quality Problem Has No Clean Fix

Three threads are circling the same anxiety: readers cannot reliably tell human writing from AI writing, tools like Pangram that try to detect it are imperfect, and the people producing AI-assisted content often know their work is lower quality but do it anyway. The 'Revolt of the Reader' thread and the 'Intellectual Fly Is Open' thread both grapple with this directly. One commenter points out the irony: a piece arguing against LLM writing contained a suspicious number of em-dashes, a known LLM tell.

The 'LLMs as a Cognitive Virus' thread frames this more starkly, arguing that LLM outputs are shaping how people think, not just what they write. The concern is that a generation of writers is training itself on LLM outputs, compressing the range of ideas and styles in circulation.

HN itself is feeling the pressure. There is an active push to flag AI-generated submissions, and the site guidelines now explicitly say 'Don't post generated text.' But enforcement is hard, and the community is aware that the problem will get worse before any solution emerges.