LLMs will exhaust human-written training data by 2027, producing widespread 'text pollution'
Author argues that human-generated corpora used to train LLMs will be largely exhausted by 2027 or sooner, after which most available text will be AI-generated or human–AI co-created. This will create 'text pollution': readers will become defensive and avoid long-form posts, creators will overuse AI and generate more low-quality content, and traditional investigative and on-scene journalism may regain attention as high-quality, human-linked content.