Don't classify. Hallucinate!

The author discusses a technique for tagging untagged blog posts with over 1,856 existing tags. Instead of feeding all tags to an LLM, they suggest letting the model hallucinate new tags based on content, then using vector embeddings to find the closest real tags from the existing corpus. The post includes an example prompt from Doug Turnbull for generating novel classifications, demonstrating the approach for a furniture query. This is a practical tip for content organization and information retrieval.