Illustrated Essays · Machine Intelligence
The intuition behind agent intelligence.
What does it mean for a system to remember, to plan, to act? These questions have precise mathematical answers. These essays find them — from the geometry of attention and the structure of memory, through model internals and agent cognition, to embodied intelligence and physical AI. Built from first principles, with original diagrams.
Featured
Series
All writing →How agents remember, plan, use tools, and coordinate. The internals of agent cognition — from memory architectures to multi-agent systems.
First essay in progress.
Transformer architecture, attention mechanisms, embeddings, and training dynamics. A systems engineer's view of how models work.
First essay in progress.
Bayesian inference, entropy, gradient descent, measure theory. The mathematical structures that make modern AI possible, built from first principles.
Embodied AI, robotics perception, sensorimotor learning, and world models. What it takes for a system to act in the physical world.
First essay in progress.
Currently thinking about
Full update →The relationship between surprise, prediction error, and the free energy principle — and whether Karl Friston's active inference framework offers anything architecturally useful for building agents that generalise beyond their training distribution. Also: what Gödel's incompleteness results actually imply, if anything, about the limits of self-improving AI systems.