Introduction to ChatXML and highly structured prompting

I have started a repo to define a structured prompting langauge - ChatXML - and collect examples of good prompts and similar ideas that use highly structured prompts.

Link to repo: chatXML: A proposal for a structured LLM prompt method

I’m not selling anything. Everything I post on this repo is under the unlicensed license so it’s free to use and abuse. My only goals are to see more structured prompting and elevate the art of prompt engineering.

Here is a quantitative analysis of chatXML. Conclusion: Useful but not a silver bullet. Many of the benefits are in compartmentalization and not performance.

chatXML/eval_test.ipynb

My assessment as well. I explored this question in depth with DrawIO documents here.

I believe Jesth may be an ideal structural approach for AI corpus development.