All right!  With all that pondering and throat-clearing done (see my previous series of posts), I was wondering what piques my interest in this LLM-hyped world from a practical side. I've been somewhat active in critiquing this whole thing over the past two years, but beyond creating AI images for the blog (or to amuse myself), or using ChatGPT to make silly little genre-busting poems (again amusement and play), and or using ChatGPT to give me a boilerplate letter that I can then tweak (marginal utility, but I guess if organization ask for things that can be boilerplated, they get something that is boilerplate).  I don't mean to dismiss the value of experimentation or play, they are valuable and low-stress ways to get to know a tool and then you may get an AHA!!! moment of a sort. I've been thinking of something more structured. 

I was listening to a relatively recent episode (it was recent when I started writing this darned post!_ of the Vergecast, titled The Chatbot Becomes the Teacher, and I started pondering... I don't like feeding PDFs or other writing into OpenAI's or Anthropic's "ooops, we just used your shit without permission" infrastructure. I also don't want to use open systems that contain a corpus of data that's not pertinent to what I am doing.  I was wondering if setting up my own GPT in a Virtual Machine to run a custom LLM, or using NotebookLM (even though I am not sure where that data ultimately goes) to mess around with my MOOC Eulogy project.  Yes... I have yet to give up on that 😂.  I have a lot of PDFs, but I'd like to start fresh with doing a sustained literature collection (articles, books, blogs, youtube videos).  Since NotebookLM can 'digest' all of that, I was thinking that it may be a useful experiment (even though I am bit uncomfortable with not knowing how the proverbial sausage is made).  Since I am pretty familiar with the literature already, having read through most of it over the last 13 years, I am wondering what sort of new "has!" might be gleaned from having all this in a custom-corpus LLM that can be prompted to give some responses.  I suspect that much of what it extrudes will need fact-checking, but I wonder if it's useful as an idea-generation or lead-generation tool. 

Thoughts?🤔