I see this idea as the overarching theory beyond personal notetaking workflow optimization, broadening into the more general problem of On Capturing Personal Data. Mainly, I stress a lot about things I want to do in terms of analysis versus the things I can actually do right now. My whole motivation for Obsidian notetaking was characterized by this incessant OCD-type discomfort around letting go of novel ideation.
Store now, utilize later focuses on quantity over quality, at least in the immediate moment. If there is an underlying latent function behind the process, then I can quell the worry that everything has to be handled right now.
This manifested in my vault through ideas like Write Stub Notes, where I now focus on “getting an idea out there” without worrying about it being ideal at that moment in time. Passive Obsidian Worktime is the security I have against worrying about utilization later, and I see the same thing happening with general quantified-self work. It is vital that I start recording data about myself, whether health metrics from wearables or At Home Bloodwork; I won’t be able to capture it later, but I can always analyze it later.
I also think about this in the context of Isomorph, knowledge management systems, and using LLMs in my vault. The conclusion I arrived at was that you can treat prompting LLMs as almost deterministic.
There is this huge rush in the AI world to figure out how to get these tools everywhere right now, because if you don’t integrate them now you will supposedly fall behind into the permanent underclass.
Where LLM content in obsidian DOES make sense though is generated transcripts of tasks done. I have a sandboxed agent memory system that writes obsidian notes about coding projects and research I do, good example was using an agent to set up my BrainAccess HALO eeg headset on homelab to record data to NeuralSet automatically for content consumption conditioning.
But with LLM knowledge bases, I see no reason to rush. How is giving the LLM my human-language, sloppy stream of consciousness now to write a formal note any different from just recording the stream of consciousness into the notes themselves, and then, in the eventual future, having an LLM utilize it? The latter seems appealing, as the tech will have settled into stable use cases and better models. That’s just it: no new information comes from having the LLM process the prompt/information, even if it looks cleaner. LLMs are great for query and ideation, but at least for the moment, the Obsidian vault should be the datastore to-be-processed in the future.
It is for this reason On Transcribing Audio Recordings is going to be a more prevelant thing going forward. I will always declare these for the most part using the callout, but getting the information on paper is the first and only priority.