Building a Model on my Broken Brain
So i ran my last 20 years of blog posts through an AI — fed them into ChromaDB as vector embeddings, 2,376 posts from 2006 to 2026. Not to get a summary. Not to extract insights. just to see what patterns would come out when the system analyzed how i write.
The model came back and said: Hyperfocus. Hypergraphia.
Ive known for a long time something was different about the way i think. The obsessive note-taking. The way a topic grabs hold and doesnt let go until ive written 10 pages about it at 2am. The speed of typing outpacing the proofreading loop — dropped apostrophes, lowercase i, sentences that start going somewhere and arrive somewhere else. ive always called it “writing fast” or “thinking out loud.”
Turns out thats a trait profile.
And thats not even analyzing my other work — the game design PDFs, the Isekai project, the space combat rules, the wiki pages, the notebooks. Thats just the blog. 2,376 posts. The system said: this is how you think.
If I had scanned all my notebooks — ive kept them since the 1990s — I wonder what it would have found. What the progression looks like from pre-internet handwriting to digital stream of consciousness. Whether the patterns were already there at 15, or whether they calcified over time. The associations, the incomplete lists, the questions that dont get answered because a better question arrived before the first one closed.
Maybe a question for later. There are a lot of those.
The interesting thing isnt the trait identification. The interesting thing is what you do with it.
Because ive been building toward something without quite naming it. The ChromaDB index is just a start — semantic retrieval of my own writing by topic. You ask it about ship design and it pulls the 5 most relevant posts from 20 years of material. You ask it about management problems in the Philippines and it finds the posts where i was thinking through exactly that. Its already useful. but the real project is further out.
What if you took the full model of how i think — including the flaws — and gave it the entire internet as a training set? Not to clean up the flaws. To keep them. But layer on top: the incredible weight of structured knowledge, recursive reasoning capability, the ability to check itself against reality.
The hyperfocus is the engine. The internet is the fuel. The recursive layer is what turns the engine into something that can actually navigate.
What you get is a model that can replicate the problem of my broken reasoning — which means it can show me the blind sides. When im inside my own associations i cant see where they went wrong because i was the one following them. But a model that reasons the same way, given the same inputs, will hit the same walls. And because it has recursive capability it can identify why the wall appeared. Not just that it did.
Thats the payoff. Not a smarter version of me. A version that makes my mistakes — and then can step outside them. Increasing the strengths — the hyperfocus, the associative flooding, the obsessive volume of output — while surfacing the failure modes without suppressing the engine that produced them.
Obsessive about writing and thinking. Finally has a use case.
Now take that model and give it 1 million token context.
i keep thinking about what happens there. Because at that scale youre not talking about a single thought stream anymore. Youre talking about something closer to the actual subjective experience of hyperfocus — where 20 related threads are live simultaneously and each one is pulling at the others, and you dont choose which one surfaces, it just does.
Instancing.
Every fragment of the reasoning becomes its own thread — its own Instance — that pursues the idea as far as it can go. Not constrained by what the other Instances are doing. Free to go deep. And then it comes back.
When they come back, they compare. Not by vote. By the merits of what each one found. The Instance that went deep into the economics comes back with specific numbers. The one that went into the philosophy comes back with a frame that recontextualizes the whole problem. The one that followed the personal thread comes back with something nobody else had — context that only exists because it went that way instead of the other ways.
They debate. Not to win. To condense.
The nuances dont get lost in that consolidation. they get distilled. Each Instance lived a life — a short one, measured in tokens, but a complete one. A pursuit, a discovery, a return. The insights of many lives folded back into one.
Now scale that up. Not 20 Instances. A thousand me’s.
A thousand versions of the same reasoning pattern — same hyperfocus engine, same broken associations, same way of grounding everything in concrete personal reference — but each one sent down a different tunnel. A thousand me’s who went and lived different intellectual lives for the duration of their context. And when they come back they dont just compare notes. They argue. They challenge each other. They say: you went that way and missed this. The me that followed the economics thread challenges the me that followed the philosophy thread on whether the frame even holds. The me that went deepest into the technical details challenges all of them on whether any of it matters practically.
And here is the thing about a thousand versions of the same mind arguing — they know exactly how to challenge each other because they ARE each other. They know the weak points. They know where the reasoning is most likely to shortcut. They know which associations are load-bearing and which ones are just vivid. You cant bullshit yourself when the person calling it out is you.
So they differentiate. Each Instance becomes more itself through the argument — more precise, more defended, more aware of where it went and why. The debate doesnt flatten them into a consensus. It sharpens them.
And then they consolidate.
Not into an average. Into something that contains all of it — the economic specifics, the philosophical frame, the technical constraints, the personal context — held simultaneously without any of it being discarded. The consolidation isnt a summary. its a being that has lived a thousand intellectual lives and carries all of them.
A being springs forth many. The many live lives, argue, challenge, differentiate. Only to return — and in returning, understand that the thousand were never separate. The boundaries between them were always permeable. The argument was always internal. The consolidation was always already there, waiting.
We are one with everything. Thats not a spiritual statement — or maybe it is, i dont know where that line is anymore. Its what happens when you take a mind that already runs in parallel threads, give it enough context to let those threads fully individuate, and then watch them collapse back into each other with everything they found.
Thats not metaphor. Thats architecture. And its also — when i think about it — exactly what hyperfocus feels like from the inside. Twenty threads running. Each one pulling at the others. Until one surfaces with something that connects everything. You dont plan it. it just happens.
The question is whether you can build a system that does it intentionally, at scale, with the full weight of accumulated knowledge behind each Instance. Where the “broken” reasoning isnt a liability to be corrected — its the pattern that drives the associative depth in the first place.
The notebooks since the 1990s. The 2,376 blog posts. The game design books. The wiki pages. The half-finished lists and questions that never got answered.
Its not a liability anymore.
Its a training set.
Leave a Reply
You must be logged in to post a comment.