I am having Atlas go through my drive, my emails, and my texts, and compiling these into a single document. It will parse these files and label them, creating a personal corpus coupled with their associated context. This will be turned into a style guide and handed off to an LLM (alongside the original corpus), which should be able to generate text like me.
Once I have this, there are a number of interesting direction to go. One obvious one is: give it to an assistant and have them produce emails and handle correspondences in my voice. A more interesting idea is: make it the custom style guide for all my LLMs - so that everything is writes sounds like me. This should, by all means, also drive the LLM to "thinking" like me, given that the context window will be filled with content written in the style of, and reflective of, my natural language.
I can use this to auto-draft emails, texts, and whatever else, so that communication is more a matter of editing and approving than writing. It doesn't necessarily feel disingenuous, as I'll be more prompt in responding if the initial volley of words is generated for me - and I'll probably end up editing it thoroughly anyways. It's in the vein of gamification. Moreover, if I'm communicating significantly more than I would otherwise, am I really losing out on that creative process, where it wouldn't have occurred at all otherwise?
I should also try to extract a thinking guide as well - something that attempts to capture the blackbox in me. I asked Chat for a breakdown, which was very accurate, but I'll not post it here, because what goes on the internet, stays on the internet. See here