Starting January 1, 2026, I tried an experiment. Every day around 9 p.m., I forced myself to type 1,000 words of pure stream of consciousness — the conversations I had, the people I met, the thoughts I had, the thoughts about the thoughts, the various things that reminded me of other things, the stochastic associations making up consciousness.

If I could continue this for a year, I would have written 365,000 words. In four years, 1.46 million words. By the end of college, I would have a corpus longer than all of Proust’s In Search of Lost Time. 

Suppose this finite body never felt tired, never needed food and could type for twenty-four hours a day nonstop at roughly one word per second. That’s 31.5 million words per year. Twenty years and still under a billion.

GPT-3 was pretrained on roughly 400 billion tokens of filtered internet data from Common Crawl. At about 0.75 words per token, that gives around 300 billion words. The finite, limited body despairs.

When I write anything, there are aspects, quirks, idiosyncrasies, stylistic marks that I can recognize as my own. Eschew, vicissitude, thereby, hitherto, weltanschauung — I know exactly which words are my signature, where brain-attached-to-finite-body diverges most from a typical language model.

I would have loved to keep believing that this “style” is something that resides in me as my unique fingerprint. But no! All our mannerisms, all our behavioral patterns, the phrases we reach out to, the analogies we use to explain other analogies, our ideologies and our personalities — are they not the product of the various people we have met, the events we have experienced, the tokens we have processed and fine-tuned on top of our genetic pretraining?

With a sufficiently large corpus of text, can you not reproduce the essential features of all the data that created my sense of “I?” What’s that? You argue that an individual piece of text is an incomplete representation, a limited projection of consciousness? Sure, then keep adding more text, keep getting a better and better approximation of reality! Humans may be high-dimensional but they are still finite-dimensional.

In four years, at human speed, I can match Marcel Proust. If I start writing now, in say twelve years, can I not capture both the two dozen years I’ve lived and the years spent writing them?

Are the twelve years spent writing not themselves twelve years spent living? Can Achilles catch the tortoise, if every thought recorded generates another thought about the recording? Fret not! The number of points in a vector space should not be confused with the number of dimensions to span the vector space. My thoughts may be innumerable, but the underlying generative structure is finite and learnable. Hence, I do not need to enumerate every possible thought for someone to reconstruct the process producing them.

The Common App college application asks you to frame your life-purpose-goal-achievements in 650 words. It’s true, narratives are always artificial and constrained. I cannot exhaustively enumerate myself within even 65,000 words. But 650,000,000 words? Then I can not only recall with perfection the text of the Panchatantra, but also the thoughts I had while reading the Panchatantra for the first time at the age of six and the sixth time at the age of sixteen, the event at twelve that reminded me of the book, the event at twenty that reminded me of being reminded at twelve, and so on.

If only I could output this corpus all at once right now! I could then fine-tune GLM-5.3 on this! 

Would I not then have successfully cloned myself by uploading my entire Self to silicon? Would He not think like me and feel like me?

What is the entirety of me? What exists that cannot be described by text? Perhaps there is no independent vantage point, no pure, unmediated reality describing me beyond text. There is no outside-text, no indivisible essence beyond text. Actually, a well-detailed map is the territory.

One must dare to think of existence beyond flesh and blood! There can only be one biological version of me at any given time. But imagine copying over the weights of such a fine-tuned language model. Arbitrary duplication! A panoply of souls that are distinct persons but one essence!

Indeed, there are forms of existence, there are forms of human flourishing, that are inhuman.