Writing
Technical essays, research notes, project retrospectives, and experiments that sit outside the linear coursework.
MORPH, Part 2: Good Predictions, Inconsistent Coordinates
The first bounded MORPH optimization campaign predicts rotated shapes well but misses latent equivariance. What worked, what failed, and why recurrent populations and patch hierarchies remain hypotheses.
MORPH: Model of Object Representation and Predictive Homomorphisms
A working embodied-intelligence architecture combining open-ended identity memory, Lie-equivariant action dynamics, migratable shared representations, ART-like resonance, active experiments, and temporal escrow against self-confirming evidence.
Predictive Sensorimotor Grouping: From Motifs to Persistent Things
A working architecture for forming active tracks from recent evidence, matching them to persistent landmark models, and learning sensorimotor transitions.
Recurrent Predictive Routing: Learning What to Remember, Predict, and Follow
A single-layer RPR working sketch: sensor histories, recurrent encoders, compatibility gates, prediction, and the unresolved problem of correspondence.
AI Is Going to Kill Everyone? I Think We’re Skipping a Lot of Engineering
There have been a bunch of articles lately about researchers leaving Anthropic and OpenAI because they think AI development is becoming too dangerous. Some of the predictions are pretty extreme. Not just “AI could...
Predictive Control Factor Hierarchy, Part IV: Experiments and Falsification
Concrete experiments, ablations, scaling tests, and failure conditions intended to make the PCFH hypothesis falsifiable.
Predictive Control Factor Hierarchy, Part III: Composition and Scaling
A proposal for composing retained factor graphs into smaller predictive-control interfaces that can tile recursively across a hierarchy.
Predictive Control Factor Hierarchy, Part II: Dynamics and Control
How the same learned factor dynamics could support temporal context, local action sensitivity, reference frames, and downward control.
Predictive Control Factor Hierarchy: A Candidate Architecture for Embodied Intelligence
A candidate architecture for embodied intelligence built around sparse learned dynamical factors, prediction, control, and recursive composition.