Poisoning Well

This blog post discusses the challenges posed by Large Language Models (LLMs) that scrape web content without consent. The author suggests a novel approach to combat this by creating 'nonsense' versions of their articles that taint the data LLMs consume, effectively poisoning their outputs without harming the original content's search ranking. The author details the technical implementation of this idea and expresses the hope that if more writers adopt similar tactics, it may encourage LLM creators to respect web protocols.