I have been obsessed with generative AI. I'd love to talk about the big topics - AGI, ASI, human survival, immortality - but my way is down in rabbit holes. So let's write some code (or just very slightly modify some existing code), and maybe we'll tackle elephants in various rooms another time.
To jump to the how-to, check out the video
Use ChatGPT API to create Knowledge Base articles in Org Mode! - YouTube ≫ www.youtube.com
Programming in natural language
One of the mind-blowing aspects of coding for LLMs is that we will be "programming" primarily in human language.
The small amount of Python code is just a little wrapper to do things like set prompt strings, pull in the API key, write files and so forth. Tiny little inconsequential things. Heavy lifting is left to the LLM to understand the source text, generate the metadata we asked for, summarize it, search it, update.
Create and search Knowledge Base articles
David Shapiro is a great YouTuber who talks about the big topics but also makes coding video when insomnia strikes. He recently did a video about using ChatGPT to create structured knowledge base (KB) articles. His version writes out YAML files. We're going to do some tweaks to the prompting and metadata to accomplish the same in Org Mode, which will then allow us to integrate the articles with a structured note taking system such as Org Roam or - in my case - the Denote package by Protesilaos.
No structure? No problem!
Our data can be as unstructured and motley as we wish - no need for a relational ERD or a schema or an ontology or a dictionary. The LLM can find the relationships for you. And for that matter, it's happy to generate SQL, SPARQL and other query languages, as well as the schemas and ontologies you might want!
So the code we will be using is a tweak on David Shapiro's code. The repository I'm using is a fork of his repo at https://github.com/gregoryg/KB_microservice, branch 'org-gab'.
Happy coding!