In the third video in my Over the Shoulder YouTube series, I demonstrate how to use an LLM with Web search capabilities to update a subset of posts on this website. Specifically, I’m looking to replace Amazon affiliate links on book notes posts with more neutral links that give preference to the book authors’ and publishers’ websites.
As with the previous video, I used Simon Willison’s llm command line tool, which allows you to leverage LLMs within traditional command line workflows. But the more “mainstream” model APIs (e.g., those from OpenAI and Anthropic) don’t search the web out of the box, so I used one of Perplexity’s models in this case.
This video demonstrates only one search. In the next one, I’ll show how to do this at scale, over all the files in the set. In particular, I’ll focus on how I’m keeping myself in the loop, which is critical given that this script’s output will be user-facing. (I.e., it’ll change my site’s content.)