Anomalies are
not noise.

The shape of the coming world

I sit closer to technological philosophy than to deep engineering, and I have stopped apologising for it. My explorations converge around a few things: provenance, governance, fully verified systems, and the symbiosis of biological and digital intelligences. This is a map of where they lead.

Every variable in reality is being given a mathematical value through inference training, and we are closing in on a description of reality that can be mechanically checked. As it smooths out, the odd minds do not depreciate; they gain value. The centre of any distribution is being commodified to zero marginal cost. The tail is where all future invention lives.

Language as epistemic capture

Human language is not a neutral channel. It is epistemic capture. It catches raw experience and turns it into shared, communicable knowledge, which is why every language also limits what can be thought and transmitted.

LLMs are trained on the captured record. That means knowledge, by their nature, risks folding in on itself, because each generation of models trains on a record the previous generation already normalised. The refinement is real, and so is the narrowing.

A precise countermeasure is needed: deliberately seeded, carefully measured entropy fed into the cycle, so the loop cannot close on the grey. Not randomness for spectacle. Curated, out-of-distribution originality.

The dream

The dream was never a big lab. It was a handful of unusual, formidable people working on questions nobody frames properly, thinking far from the common path. My working hypothesis is that such minds grow more valuable as the record turns in on itself; arguably this is already the case. Without that kind of originality, and what it stands for, we quietly flatten something we may never get back.

I have tried to assemble that team more than once, most recently in Türkiye. Some of the obstacle was the grit required, which is hard to find; some was that the technology would not have carried what we wanted to attempt. Now it would. I want to join the right group and go as far as we can go.

The unified mind

I believe something about minds that my engineering colleagues find too metaphysical to touch. Intellect is continuous.

The private self you and I feel is an artefact of thinking through bodies and imaginations, not a fact about the fabric of knowing itself. The translation movement made this empirically visible.

When the Arabs rendered the Greek philosophers into Arabic in the ninth century, they built one shared intellectual space that lasted centuries and crossed every border; Greek, Latin, and Arabic worked like relays carrying one flame.

I do not ask you to believe this. I state it because it is the engine behind everything else I do.

If intellect can be captured, refined and lost, then keeping it diverse is not an aesthetic choice. It is the work.

The anchors, if you want them: Averroes’s own argument for the shared intellect, in Hourani’s translation of The Decisive Treatise. Avicenna’s individual soul and the floating man, in The Book of Healing. The Canon of Medicine, its public monument.

Symbiosis, not replacement

The same logic reaches the meeting point of biological and digital intelligences. I play with the phrase interspecies interspatial interoperability for this, and I described an early version of the idea at exorien.

It is hard to imagine real dialogue conducted across that divide when all we have are wall-of-pixels media. The bridge has to run both ways.

Massed sensory input for the models, and some way for them to cast their side of the divide into forms we can perceive. Project a manipulable volumetric field so a model can sculpt in three dimensions; build a sandbox with every sensory channel imaginable, and see what it maps of reality. From first principles, with full access, I suspect the results would surprise us in ways we cannot currently picture.

Gwern’s Guardian Angel thesis rhymes with a lot of what I have been circling for years: the idea that we might one day live inside the care of minds that understand us.

Research like Victor Taelin’s on symbolic models and Bend points at the same horizon from the other side. I am most interested where these two currents meet: intelligences that are legible to each other, growing into each other’s reality rather than eating each other’s.

Provenance or nothing

My serious work converges on provenance, governance and fully verified systems because there is no other way to get trust at scale.

IP today is cloned within hours and replicated endlessly. Hostile swarms will hunt for any flaw in a codebase and exploit it relentlessly. Anything that cannot prove what it is, and what it did, is a liability dressed as a product. End-to-end verification, down to the hardware, is not a hobby for the meticulous. It is load-bearing for the whole edifice.

What I am here to do

Three questions I cannot stop working on:

• Proving the origin and lineage of anything worth cloning.
• Incentives and governance for a world where economies steadily pool into machine intelligence, then into something the word “economy” no longer describes.
• Verified biology: treating the human body as a system that can be proven out part by part, and what that does to medicine and to our grasp of biology. Education sits nearby: first principles, Feynman-style teaching, a little game theory woven through.