flyputer
Local Gemma agent queries FlyWire, runs small LIF subcircuits, and includes a playable Giant Fiber swatter game against the real LPLC2/LC4 escape reflex.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.The simulation engine is a current-based LIF with stated v_th, tau_m, tau_s and refractory period.
https://github.com/migkapa/flyputer/blob/775da29578880e77d9eeb7fc3a8dc039b970673c/flysim.py#L177-L208The claimed dopamine-gated depression is a boolean OR into a per-Kenyon-cell mask, not a write to any synaptic weight.
https://github.com/migkapa/flyputer/blob/775da29578880e77d9eeb7fc3a8dc039b970673c/sniff.py#L83-L98The KC->MBON weight matrix Wkm is constructed once from connectivity and is never modified afterwards.
https://github.com/migkapa/flyputer/blob/775da29578880e77d9eeb7fc3a8dc039b970673c/sniff.py#L58-L63
Real FlyWire subcircuits run through a vectorised current-based LIF engine with stated threshold, membrane and synaptic time constants and a refractory period; the compass, escape and optic-relay demos all use it. Not L2 despite sniff.py advertising 'a dopamine-gated DEPRESSION of KC->MBON output': the KC->MBON connectome weight matrix Wkm is built once in circuit() and is never written to anywhere in the repo, and the 'depression' is a boolean per-Kenyon-cell mask updated by `self.depressed |= code`. The score it produces is the fraction of a cell's KCs in that set and never multiplies through Wkm at all, so no synapse in the modeled circuit changes value. This is a sparse-coding demonstration, which the module's own docstring concedes ('demonstrates the architectural MECHANISM'). Also worth noting: the mask is monotonic and never recovers, so it could not express extinction or forgetting even in principle.
Record
Nearby
All projects →MaleCNS simulation connected to ViZDoom through modeled visual inputs and a fixed button readout. Includes plasticity experiments and negative validation results; learned survival has not been demonstrated.
Research framework comparing a MaleCNS-constrained sparse recurrent controller with rewired graphs, conventional neural networks, and other controls on simple VizDoom tasks.
Ratings on thelearningfly.com are proposals, not verdicts. Every one of them is arguable in public.