Ouroboros

10/14/2025

First you take all the people that have ever been alive in the last, let’s say, 4,000 years, when the glyphs were first assigned sounds.

Then you take that subset of those people that could be bothered writing. Over that 4,000 years that bar has dropped somewhat. But still, it’s a premium slice of humanity, from a cognitive perspective.

Then you train a machine to engorge all that writing; all of it. That machine can then use that material to answer any question, comments on any thought, etc, based on interpolation of all that learning material.

There’s two steps to this. Step 1, which word follows which? Step 2, which information sources to use?

Step 1 is pattern prediction: a model trained to guess the next word based on probability distributions from all the text it’s eaten. It doesn’t know meaning; it just knows what tends to come next when a human sounds like they know what they’re talking about. That’s the interpolation layer, the statistical meat grinder doing its thing.

Step 2 is source selection: deciding which words, and therefore which contexts, to draw from. That’s where retrieval, ranking, and curation creep in; what the model reads, remembers, or ignores. It’s a bureaucracy of algorithms.

The first step builds fluency. The second step builds logic. Together, they can create the best and worst of every human expert committee.

That’s the gist of it. Humanity spends four millennia bleeding words onto clay, parchment, paper, and screens. Then we pour it all into a statistical meat grinder that figures out how one word tends to follow another and call the result artificial intelligence.

It sort of is too. I can prove that by noting that a small fraction of modern writers are claiming copyright theft by the owners of the LLMs. The intelligence bar is pretty low.

These machines don’t think, they interpolate between every philosopher, crank, mystic, blogger, unpaid intern, and copyright theft victim that ever strung a sentence together.

What comes out is an odd sort of weighted average of human written thought on any subject. Which in itself is compelling because previously we never knew there was such a thing.

Since, from here on, these machines will be used to write everything, arguably humanity has just stopped progressing. Ongoing ingestion won’t add anything new – we’ll just get into a recursive loop. The IT equivalent of Mad Cow disease.

This is either the beginning of the end or the start of the future, a very mediocre one where nothing genuinely new ever occurs.

My suggestion is that we engineer an innovation capability into the machines before we forget how to innovate.