The Lab · 0 of 8 live
Learning, representations,
dynamics, computation.
Eight small experiments, each built from scratch, each asking one question, each running live in your browser. Together they are the ladder that leads to Reverie.
One question
Isolate a single mechanism and ask when it appears or breaks.
A scaling ladder
Three or four points along one axis: does the effect hold, grow, or vanish?
Honest limits
Every write-up ends with what would still need testing at scale.
Runs in your browser
Trained in Python, exported, and run live on this page. No server, no waiting.
- 01LearningUp next
Search Landscapes
Gradient descent, CEM and evolution strategies: does their ranking flip as dimension grows?
- 02ComputationPlanned
Hopfield Memory
Why is attention a Hopfield network? Classical versus modern storage capacity.
- 03DynamicsPlanned
ODE vs NN
Does a Hamiltonian inductive bias fix extrapolation and energy drift?
- 04RepresentationsPlanned
Latent Observatory
Which training objective recovers the true generative factors?
- 05LearningPlanned
Local Learning
How fast does the gap to backprop grow with depth?
- 06DynamicsFlagshipPlanned
Neural Cellular Automata
Does growth trained by evolution strategies regenerate differently from growth trained by backprop?
- 07RepresentationsPlanned
Grokking
How do weight decay and data fraction change which circuit forms?
- 08DynamicsPlanned
Tiny World Model
How long does the latent keep an object it can no longer see?
- →Research
Reverie
The full study: does a world model represent what it cannot see, and does it use it?