I've written some blog posts before, but I wouldn't say I'm a natural at it. It's not easy for me to churn out articles, I just don't have that much time to sit down and structure everything the way I want.
I came across Jake Brukhman's article on LLM pseudolanguages and started wondering if it could help me write articles faster. After a bit of tinkering, the result is a markup pseudolanguage called Prose Markup Language, or PML.
PML is a domain-specific markup language that simplifies the process of writing articles by providing predefined tags for different elements, such as introductions, conclusions, paragraphs, and ideas. In addition, PML includes style header tags that allow you to give hints on the desired article style, such as "write from a first-person perspective" and "use simple to understand words."
This new language enables people who aren't naturally good at writing or don't have the time for it to quickly generate an article from their insights. PML streamlines the editing and revision process by allowing users to easily rearrange, add, or remove sections of their article without having to rewrite large portions of text. There's value in taking disjointed thoughts and weaving them into prose that flows and is easier to read. With PML, you can just do a brain dump, bad grammar and all, and it gets turned into organized prose.
But PML isn't just a markup language; it's a writing collaborator. When I first started working on my PML script, I included generic instructions like "add two other benefits to the user if they use PML." The benefits that ChatGPT generated were interesting enough that I actually included the specifics of one in my final PML script. This process is a lot less dry than trying to write an entire article by yourself. The article becomes the final result of having a conversation. PML provides a collaborator that you can explore ideas with, brainstorm, and test ideas against. The iterative nature of PML allows you to specify basic instructions, see the output, and refine those instructions to achieve the desired result. PML also lets you experiment with different writing styles to find what you like best.
I'm already surprised by how good PML is; it's much better than I was expecting. It's exciting to imagine what later versions of GPT will be able to generate. If you're interested in trying PML for yourself, you can find the specification on GitHub at https://github.com/dineshraju/pml.
(This note was generated with PML. You can find the original GPT generated article here and the PML used to generate it here.)