Really nice simple to understand article. I do wonder if the capabilities demonstrated here are specific to the LLM in use in the article or would most models display similar behaviour?
It's a very elegant idea and the article did a great job at introducing it and explaining how to use it.