Infinite Sugar
Browser artwork placing a FlyWire-based fly in a terrarium with continuous sweet-sensing input. Neural activity drives some movements while wings and small foot motions use supplied patterns.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The kernel header states the model, the dataset size and every LIF constant used.
https://github.com/cnqso/infinite-sugar/blob/fdbbd866b203a709a164970b0a8996108edd57a5/web/brain.ts#L1-L21Weights are unpacked and scaled once at load into a Float32Array handed to the Brain constructor.
https://github.com/cnqso/infinite-sugar/blob/fdbbd866b203a709a164970b0a8996108edd57a5/web/brain.ts#L137-L142The propagation loop only reads w[k] to add to the postsynaptic voltage or the inhibition queue; no weight is written.
https://github.com/cnqso/infinite-sugar/blob/fdbbd866b203a709a164970b0a8996108edd57a5/web/brain.ts#L216-L228
A whole-brain leaky-integrate-and-fire run over FlyWire FAFB v783 - 139,255 neurons and 2,700,513 edges at 1 kHz, with a 20 ms membrane tau, threshold 1.0, 2 ms refractory and a 4 ms delayed-inhibition queue, all stated at the top of the kernel. L1. Not L2: the weight array is unpacked and scaled once when the blob loads and is thereafter only read inside the propagation loop; there is no plasticity, dopamine, or training code in the TypeScript sources at all. The sweet-sensing input is continuous stimulation of gustatory neurons, which changes voltages and nothing else - the piece is explicitly an artwork about that, not a learning claim.
Record
Nearby
All projects →Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.