The revolt of the reader

The writer expresses frustration with the increasing prevalence of content generated by language models (LLMs) being presented as authored by humans. They argue that readers can easily detect LLM-generated text, which disrupts the trust between writer and reader. Readers explicitly avoid LLM-authored content and writing by authors they respect, even if the work is well-written, because they prefer human-created pieces with all their imperfections.

The writer contends that using LLMs to write undermines the social contract between writer and reader, as readers are now expected to sift through LLM-generated content to discern what is real. The writer compares the issue to the rise of email spam in the early 2000s, predicting that a similar solution may emerge to address the problem of LLM-generated text.

They mention that Pangram Labs has developed a model, Pangram 4, which they find to be highly accurate in detecting LLM-generated text, and have implemented a policy requiring the use of Pangram 4 to verify the authenticity of public writing.