Simon Willison describes Doug Turnbull’s technique for tagging blog content without constraining a model to a fixed vocabulary. Rather than selecting from an existing set of 1,856 tags, the approach prompts an LLM to freely generate novel tag suggestions for a piece of content. Those generated suggestions are then matched back against the real tag corpus using vector embeddings to find the closest concrete matches, using the model’s free-form guess as a semantic anchor rather than a final answer.
