Julia forum debates auto-deleting LLM-written posts

Julia’s Discourse forum is arguing over whether to automatically detect and remove posts generated by large language models, after a longtime participant said a new-package announcement had become “hundreds of words of LLM slop” he found unbearable to read.

LeePhillips urged the community to follow other technical fora that ban machine-written comments outright, citing detectors that claim more than 90 percent accuracy, and warned he would leave if the practice spreads. He wants questions and announcements written by people.

Others rejected a pure origin ban. Several argued the real problem is low-effort, meandering text that forces readers to dig for signal, a habit that existed before chatbots and is already covered by the site’s “be concise” expectation. Mason suggested tightening enforcement on length and usefulness rather than carving out an LLM-specific rule, noting that purely human walls of text once sparked more fights because people actually finished them. Detection of origin, he and others implied, is easier than judging quality, yet quality is what readers notice.

A separate line of argument stressed equity. Solo developers and non-native English speakers may lean on models to match the polished output of larger teams that already enjoy communication advantages; forbidding the tools, some said, entrenches that imbalance. Counter-replies insisted that obvious machine prose still damages the poster’s reputation and that imperfect human English (or careful short posts) remains preferable to dense, audience-blind filler. Proposals to allow native-language posts drew quick pushback on the grounds that they simply shift translation work onto every reader.

Moderators restated that the board is moderated, threads stay on a single focus, and meta complaints about censorship are themselves off-topic. The exchange leaves the community weighing automated filters against sharper, more subjective policing of verbosity, with no policy change announced.