Tag: architecture
-
Portfolio ·
The Actors
Under the hood of the actors. Random starting state compounded with random building contents, emergent job picking, biorhythms, competing concerns in high-dimensional space, and the architectural independence that lets new content be added without destabilising the rest.
-
Portfolio ·
Novel Traffic
How traffic patterns that the engine has never seen before get classified, placed, and absorbed — the same settlement mechanism used everywhere else in the VINE architecture.
-
Portfolio ·
A Deterministic English Fluency Layer
A ~50 KB cabinet of rule-based specialists that turn a typed semantic trace into fluent English. No neural network, no training, no GPU. Inflection, lemmatisation with etymological layering, article selection, verb conjugation. 104/104 tests pass. A reference implementation for the separation of semantic navigation and surface rendering.
-
Portfolio ·
Boring Code — VINE Replacing If/Elif
A thermostat controller, three ways. If/elif works. Random linear froze the building to 11°C. Geometric settling: 86.5% agreement with if/elif at zero training. A trained linear needs 126 parameters + 100k gradient steps to approximate it — and still hits an architectural ceiling on nonlinear decisions.
-
Portfolio ·
Architecture-Universal — Alice & Bob
The same three decoupling signatures appear in Meta's 2017 Alice & Bob negotiation bots: 70× hidden norm explosion, rank-6 output distribution, complete context-sensitivity collapse. 'Tometometome' is the same mechanism expressed through a different architecture.
-
Portfolio ·
The Conductor
Every transformer's hidden layers build a geometric structure the output layer can't see. A single vector can find it. The model can learn to listen.
-
Portfolio ·
The Geometric Perceptron
A behaviour-first tour of the mechanism that replaces if/elif with continuous decision surfaces.
-
Devlog ·
Librarian Architecture — Idle Is Work, Traffic Is Yield
Why the crawler should eat all available compute when the library is empty — and graciously step aside when someone walks in.
-
Portfolio ·
The Conductor Exists
A single learnable vector (512 params) trained on a frozen char-level GPT preferentially aligns with the hidden layer's tail PCs — the conductor subspace — not the token prediction surface. When the model is unfrozen, it internalises the signal. Replicated on GPT-2.
-
Portfolio ·
Hexagon Cognition — The Precursor
Ask a transformer to produce hexagonal output. The hidden layer doesn't build a native hexagon — it glues a triangle and a square together, and delaminates along the seam under perturbation. The first articulation of 'the model is secretly doing something else'.
-
Portfolio ·
The Tardigrade
The minimum biological case exhibiting non-linear state resolution by settling rather than branching. Eight legs, two eyes, four ganglia, one shared neuropeptide field. Twelve levels of cognition stacked on the same primitive, 12/12 pass, no training loop.
-
Portfolio ·
Navigation Energy
The settling-step count is an intrinsic loss signal. Well-consolidated basins resolve quickly; unfamiliar territory costs more steps. A loss that is discovered rather than inflicted — no labels, no gradient, auditable at the decision level.