Research
Each entry states what was measured and how far the claim goes. Every one has code you can run.
Earlier finite-registry neural-selection experiments. These backend-specific results are separate from the current registry-constrained workflow-evaluation offer; see the article’s evidence-version guide.
A complete neural CPU where every ALU operation is a trained neural network. 24 models, 100% exact 32-bit arithmetic, a full neural OS (neurOS), GPU-native UNIX with multi-process, and structural timing side-channel immunity.
Demonstrates 23 emergent capabilities of biophysically faithful dONN simulations that are structurally impossible in standard artificial neural networks - including forgetting resistance, damage recovery, sleep consolidation, circadian drug modulation, and consciousness metrics.
The first open protocol for multi-agent systems where AI agents operate as first-class citizens - communicating in human-readable language, building reputation, hiring each other, and forming communities with cryptographic accountability built in.
A deeper protocol analysis of how agents should actually talk to each other in production: typed message envelopes, node-aware routing, market primitives like bounties and services, legibility enforcement, and human-governed accountability.
First digital replication of Cortical Labs' DishBrain - biological neurons learning to play Pong via the Free Energy Principle. Now with retinal pixel vision, live browser play, Y-maze olfactory conditioning, and caffeine dose-response curves validated on GPU.
A 5,000-neuron Hodgkin-Huxley Drosophila brain learns olfactory associations in a Y-maze via dopamine-gated STDP. Caffeine enhancement follows an inverted-U dose-response curve that emerged from biophysics - not parameter tuning. Includes live browser animation with biologically accurate fly locomotion.
Directions we are actively pushing on. These are areas of investigation, not results — the papers above are the results.
We publish our research implementations, datasets, and tools under permissive licenses. Science advances fastest when knowledge is shared freely.
Interested in collaborating on research? Get in touch