INTERACTIVE RESEARCH

Research you can
play with.

A place to explore intelligence by watching it learn.

Step inside the experiments behind the work. Start a task, follow its decisions, and see what changes as a model learns.

Launch Polymorphic Play
EXPERIMENT 01BROWSER DEMO

Polymorphic Play.

Four minds. One little world.

Watch v1, v2, v3 and v3 adaptive grow real neural networks while reconstructing a changing 3D heat field. Follow the shadows, confirmations and committed experts in one guided scene.

Go + WebAssemblyLive Three.js sceneRuns in your browser
Open Polymorphic Play

Opens full page in a new tab and starts automatically after downloading a separate engine of about 9.3 MB. Its default story takes roughly three minutes at 1× pacing, with pause and speed controls.

Polymorphic Play with arrows identifying the neural network, correct target field and model prediction.See the network. Follow the learning.

A CONTROLLED WORLD

Explore the behavior, not just a score.

This is a procedural learning demonstration with an analytic target. Model size, search budgets and update counts stay visible so you can compare the four policies in context.

Explore the Welvet engine