Training Wheels for My Mind
Riding Wobbly
I thought my AI-powered workflow would be a smooth ride. It wasn’t.
Last week, I talked about AI as a bicycle for my mind—a tool to speed up learning and thinking. But if you’ve ever learned to ride a bike, you know the truth: before you go fast, you go wobbly. You scrape your knees, crash into bushes, and question why you ever thought this was a good idea.
That’s exactly how my AI note-processing feels. Exciting, frustrating, unpredictable—and occasionally, hilarious.
Starting Small: Training Wheels On
I know better than to go full speed from the start.
Instead of dumping decades of notes into an AI and hoping for magic, I started with a small-scale proof of concept. A handful of manually loaded notes. A simple test: Can AI turn my raw notes into clear, structured insights? (I call them “Atoms”)
I wanted the AI to summarize the content of each note, tag it so I could find it later and make it as short and clear as possible. Here’s an example note:
Desirable Difficulty
Learning sticks better when it’s effortful. The harder your brain works to retrieve or process information, the stronger the memory and understanding.
Struggle strengthens recall. Easy learning is often forgotten quickly.
Example: Spaced repetition and active recall force your brain to work, improving retention.
Origin: Research by Robert Bjork, expanded by Barbara Oakley and Scott Young.
Tags: #Learning #Memory #CognitiveScience #Education #Resources
It’s more than just a definition because I want to capture the full meaning, but it’s much shorter and more easily scanned than the original note.
Early signs that I had set up the note processing right were promising—until I hit the weird notes.
“Dentist 4/11 at 3” → Is this an insight or a calendar reminder? Why is it even in my notes?
“Humidifier filter: HCM-350” → Summarize this? Into what? But wait, this is actually one note that I need to save! Have to tell the AI when it runs across notes it can’t figure out to flag them for me to review and either keep, revise or delete. (Make a note to do that!)
“Grocery list: coffee, eggs, batteries” → The AI confidently summarized this as “Essential consumables for daily sustenance.” Thanks, but no.
Clearly, not everything needs to be atomized. Some notes belong in a task list, others need context for AI to process correctly. Some needed deleting and quite a few came up that I couldn’t imagine what they meant.
This wasn’t just a test of AI—it was a test of how I actually use my notes.
The Wobbly Ride: Crashes, Overcorrections, and Surprises
I thought AI would smooth out my workflow. Instead, it made me rethink it completely.
At first, AI over-summarized everything—reducing rich insights to dry bullet points. Then, it went the other direction, rewriting my thoughts with WAY too many words.
Even defining an "atom"—the core unit of knowledge I wanted AI to extract—was tricky. Some notes worked perfectly:
“Loss aversion: People fear losing more than they enjoy gaining.”
But others got murky fast:
“The way to help India is not to ask how to help, but to ask what does the most damage?”
AI flattened the nuance, missing the real insight. What it’s supposed to mean is that you look for the things that do the most damage and tackle them first rather than try to boil the ocean and fix everything a little.
This was not a plug-and-play system. I had to fine-tune, adjust, and rethink the process—just like learning to ride a bike.
Finding Balance: The Highs, Lows, and Sudden Stop
Then I found the sweet spot.
After weeks of trial and error, ChatGPT-4.5 nailed it. It generated concise, structured atoms with perfect summaries, great tags, and just the right balance of brevity and depth.
Finally, it felt like the training wheels were coming off.
Then—I hit the paywall.
I wasn’t on the Professional plan at $200/month (I’m on the monthly user plan at $20), and I was about to run out of responses. The perfect system I had built wasn’t scalable.
But instead of starting over, I captured the process. I asked ChatGPT 4.5 to write out a prompt for other AI models to use. It built a detailed prompt that would let any AI atomize my notes. Training wheels or not, I had something solid to work with. I still have to copy and paste notes into the AI, but this is progress.
Leaning Into the Ride: Unexpected Lessons
Ironically, manually loading notes is a feature, not a bug.
Going slower forces me to engage with my own ideas—to rediscover old insights, rethink assumptions, and spot connections I’ve missed.
Progress isn’t linear. There’s real value in a wobbly ride. Desirable difficulty—a concept backed by learning experts like Barbara Oakley and Scott Young—suggests that the harder something is to process, the more deeply it sticks. If it’s hard to do, you remember it better.
I’m not just organizing notes. I’m tuning up my brain.
The training wheels aren’t off yet, and that’s okay.
I’m still converting decades of notes. The system is still evolving. But I’m learning that the wobbles are where the fun is.
Progress isn’t about perfect automation—it’s about leaning into the process, adjusting as you go, and enjoying the ride.
That’s My Perspective