Journaling for the AI future
I was listening to a Lenny podcast on my run today. Lenny is an optimistic and prolific PM-centric podcaster that always has engaging guests that give me something to gnaw on regarding product building and delivery. While the guest I was listening to today was fantastic, there were moments when I found myself already saturated with the content, and not what I was optimally geared towards listening to or learning in that moment. Not that the time wasn’t well spent, it absolutely was and I had a few good takeaways, but in an attention economy where i likely only allocate 1-2 hours of podcasts a week, I felt like i could do better.
So then I thought, if LLMs had perfect knowledge of me and I was looking to maximize my education hours, could I ask them to pre-screen a handful of podcasts based on the transcripts, and ask which of the podcasts I would likely find most valuable from a learning perspective?
A current solution for this approach is querying LLMs that have memories. ChatGPT can be configured to store and summarize your converstaion history over time and draw on it as needed. Here is an excerpt from mine:
This is all well and good for right now, but has two big drawbacks, it’s spotty and locked in. I like using the best tool for the task, whether that’s ChatGPT, Claude, Cursor, v0, etc.., and many times the queries are contextual (Project A vs Project B, and obviously would like to avoid problems akin to when your spotify wrapped indicates your most loved album is the Frozen soundtrack, yes i’m looking at your other parents).
The solution for this is one of my favorite principles of the past few years that I’ve mentioned before, File Over App.
Applying this, and given the data isn’t quite free enough and we’re not at AGI quite yet, I’ve decided to amp up my journaling intentions heavily this year. Mundane or poignant details, otherwise not captured or externalized, will be externalized into markdown and tags, all in Obsidian, for probably not that distant future consumption. Until we have Neurallink, this is the next best way to ensure I can get truly personal outcomes, and own the data.
For instance, here is an excerpt from today;
Listened to Lenny podcast today, with the guy from Gong. It was generally good and the guy knew his stuff. Talked about his pod approach, and how they work with design partners. Other than the design partner bit, i found the “pod” approach being spoken as though it was a brand new innovation, yet it’s really just a rejig/shaping of stream aligned teams, and then from scrum and agile. He did mention Marty Kagan books of which i’ve read zero, but didn’t tie anything together. It irks me when people talk about things as being the first and brand new, but don’t pay at least a bit of homage to systems of before. Maybe i missed it, still irked tho. …
And if that was paired with another entry about how I really enjoyed something about some cool Data Science topic, or a delightful episode of Sharp Tech where I reacted more emphatically, and also paired with trends of learning and my interests, I bet it would have recommended a different podcast.
And in the off chance this makes sense; I’d like to think of my journal as a really large .cursorrules file to help with personal decision making when LLMs can perform tasks I clearly cannot (like looking ahead to tell me what I’ll likely vibe with the most). A .liferules file if you will.
What other outcomes might there be? I obviously don’t know yet! But i do know i’d prefer to have this data in a file, outside my brain, not locked in to any one vendor or model, ready for ingestion, summarization, and then personalization with “AI Next” at a scale that will really help me optimize my time, and life.
I’ll be ready for you, GPT-X, AGI-mini, or whatever is coming next.
This may be turning into a micro series, definite similarities with another piece that was focused on technical documentation.