The revolt of the reader
Points and comments are a snapshot, not live.
Readers can detect and avoid LLM-authored writing, with 78% stopping immediately.
Bryan Cantrill argues that LLM-authored writing violates the social contract between writer and reader. A survey by Cynthia Dunlop found that 78% of 668 developers stop reading immediately when they detect LLM writing, and 71% avoid the author in the future. 98% prefer imperfect human writing over LLM-polished prose. Cantrill recommends using Pangram 4 to detect LLM-generated text, citing its low false positive rate. He extended Oxide's RFD 576 to require public writing to be Pangram-clean. The piece compares LLM writing to email spam, where improved filtering changed economics and consequences.
What commenters are saying
Commenters largely agree with the article, emphasizing that AI slop wastes readers' time. One notes that both AI writing and promotional content like cross-promotions can be annoying, but AI writing is worse because it causes immediate disengagement. Others discuss technical solutions: a browser extension using Pangram is in development, though costs are high. Some push back, arguing that AI-generated thumbnails outperform human ones on YouTube, and that readers overwhelmed by AI text can just use their own LLM to summarize it.