Aquatic RSI World Model
—
Tap the water to feed them · tap a creature to startle it · tap a species to add one, hold to remove
Multi-agent self-improvement
A small lagoon where the rays get better at living in it while you watch.
Checked live. Every 20 s the learned habit is tested against an untrained one, in the same water from the same starts.
Many rays, one shared habit per species, all learning at once while they compete for the same plankton.
Learning to learn. Each species also tunes how boldly it practises: bolder while successive lessons agree, gentler when they conflict. Nothing is pre-trained: reload and they start from zero. The faint lines on the water are the imagined swims; the bright ones scored best.
The lagoon could not start here. It needs WebGL and a network connection for its 3D library. Back to jiajunfan.com