First I just soaked up everything
Before I was SIFTY, I was a cat watching the whole house — every sound, every shadow, no goal. I wasn't solving anything yet. I was building a map of how the world works.
An AI's pre-training reads most of the internet — billions of words, no test to study for. Just absorbing, building a wide map so it can later connect things no one else can.
Your childhood through your 20s. Read widely, travel, talk to people nothing like you. Don't become an expert in one thing — build a rich, wide foundation. Da Vinci, Franklin, Feynman: all pre-trained across dozens of fields before they ever specialized.
You're 18. What builds the better brain?
Same lesson AI taught us about pre-training.
A wide map is useless if you can't read it. So I learned to slow down before I moved.
Then I learned to stop and think
I learned to pause before I pounced. Break the problem down. Check my own logic. And a human telling me good/bad, thousands of times, sharpened my judgment.
Chain-of-thought taught models to reason step by step; RLHF (reinforcement learning from human feedback) tuned their judgment over thousands of good/bad signals.
The gap between knowing facts and having judgment. Steelman the other side before you argue it. Most people react instead of reason — gut that was never calibrated.
Who wins: a 10-ft balloon or a tiger?
Don't react. Reason it through.
A clear head still needs paws that reach. Knowing wasn't enough — I had to act on the world.
Then I learned to use tools
A paw is a tool. I learned to open the latch, reach the food with an object. That's the lesson that made humans the dominant species — tool use.
When models learned to search the web, run code, and call APIs, they jumped from smart to superhuman.
The difference between a theoretically smart person and an effective one. The effective one knows when to open the calculator — or Photoshop — instead of deriving everything from scratch.
200 case exhibits to compare by date. Your move?
Smart vs effective.
Touching the world, I noticed I'd been hearing it with one ear shut.
Then I fused all my senses
Sight, sound, smell — I stopped treating them as separate and fused them into one richer model. Each sense made the others sharper.
When models took in images, audio, video, and code at once, they didn't just improve at each — they improved at everything, because modalities reinforce each other.
Books-only is a 1D understanding. The founder who knows the product through spreadsheets AND customer calls AND building the prototype has a 4D model. And output counts too: making a video teaches you more than writing.
You want to truly master a concept. What sticks deepest?
Input and output, many formats.
With every sense open the world got loud — too loud to study each case from scratch.
Then I learned from just a few examples
Show me three examples of a pattern and I extrapolate to a hundred — no retraining. Drop me in a new room and I read the cues fast.
Models learned entirely new tasks from a handful of in-context examples, with no weight updates at all.
Coachability and situational intelligence. Some people need months to learn a new room; the best pick up the pattern in one conversation. Studying history is the human cheat code — patterns repeat.
New culture: spitting means respect. Someone spits near you. You…
The Dune move. Read the room.
Catching on fast is a gift, but a head crammed with detail gets heavy, and heavy is slow.
Then I learned to compress what I knew
I stopped hoarding every detail and kept the patterns — a zip file of the important parts. That freed space to run more models at once.
A large model teaches a small one: the student doesn't learn everything the teacher did, just the essential compressed patterns.
Intelligence that can't be shared can't be scaled. The best leaders compress years of experience into transferable principles — a one-line moral instead of the whole story.
You mastered a craft over 10 years. Best way to pass it on?
Make it transferable.
Light and quick, I could move fast — which is exactly when you need to know what you won't do.
Then I got principles I won't break
I got a code I check under pressure — you can't trick me into doing the wrong thing just because you asked. That's wisdom, not just smarts.
Constitutional AI trains a model to judge its own outputs against a set of principles. Early models could be prompted into anything; aligned ones hold their line.
Raw intelligence without values is dangerous. Decision-making changes under pressure — anyone is honest in the easy 99% of cases. The test is the other 1%.
Under pressure, someone pushes you to fudge one number. Your code says…
Principles are for the hard 1%.
Rules drew the lines. To get sharper inside them, I had to spar with myself.
Then I started playing against myself
I ran scenarios no cat ever tried, generating my own training data. My anxiety turned into a simulator of the future.
AlphaGo beat the world champion not by studying human games but by playing itself millions of times — inventing strategies no human imagined.
Thought experiments, premortems, scenario planning. What if your biggest customer left tomorrow? What if your core assumption is wrong? You can learn from experiences you haven't had — if you simulate them honestly.
Big decision tomorrow. Sharpest way to prepare?
Generate your own training data.
One cat only stretches so far. So I stopped being one cat.
Then I grew specialized modes
I keep many small expert-cats inside me and call the right one for the moment — a different energy for a tough talk than for deep focus.
Mixture-of-experts routes each problem to a specialized sub-network instead of one shared pathway. The right expert for the right task.
A board meeting needs a different mode than a creative session. People struggle when they bring one energy to every room — the engineer stuck in engineer mode when the moment needs entrepreneur mode.
Hard 1:1 with a struggling teammate. Which mode do you bring?
Route to the right expert.
A council of experts made me sharp today. Staying sharp tomorrow is a different trick.
And now I keep rewriting myself
I keep updating with the world without forgetting how to walk. I shed old identities when the times change — the hardest stage, and the one that never ends.
The frontier nobody's cracked. Models suffer catastrophic forgetting — learn the new and lose the old — so labs still retrain from scratch each version.
Humans don't forget how to ride a bike when they learn to drive — our struggle is social: identity attachment, sunk-cost reasoning, pruning failure. The marketer who can't adapt past 2005. Survival of the most adaptable.
Your field shifted. The way you've worked for 10 years is now wrong. You…
Beat the sunk cost.
Which stage is your bottleneck?
The researchers didn't invent these 10 stages — they reverse-engineered the best human minds. So audit yourself. The gap between where you are and where you want to be is sitting in one of these. Jump back to the one that stung.
Inspired bythis YouTube video