Fly Dino
Original Chromium Dino controlled by a fixed 80-neuron MaleCNS circuit with simplified leaky tanh dynamics and a 243-parameter readout trained with the cross-entropy method (CEM). Includes live computed activity, reproducible training, held-out benchmarks, and circuit-silencing controls. Circuit subset: only the artificial readout is trained.
Neurons are simulated — leaky integrate-and-fire or better — with documented, tuned parameters.
Dispute this rating →The evidence
No evidence, no level.Eight engineered observations feed 80 fixed MaleCNS cells with signed normalised leaky tanh dynamics; only the 243 parameters of the artificial readout are trained, by cross-entropy method.
https://github.com/cobanov/flyjump/blob/c08c86bc18efd8125964b1d2ca4fc1df59700f30/README.mdThe author states explicitly that no biological fly, muscle simulation or internal synaptic plasticity is claimed.
https://github.com/cobanov/flyjump/blob/c08c86bc18efd8125964b1d2ca4fc1df59700f30/README.mdA silenced-circuit control is reported: the same trained readout with the circuit silenced completed 0 of 100 held-out courses, against 99 of 100 intact.
https://github.com/cobanov/flyjump/blob/c08c86bc18efd8125964b1d2ca4fc1df59700f30/docs/experiment.md
Unusually careful about its own limits, and rated on what it claims rather than on what a headline might. L1 rests on signed normalised leaky tanh dynamics with published equations — note that this is a rate model, simpler than the leaky integrate-and-fire the level describes, so an L0 dispute would be reasonable and is worth having. Not L2: the author states no internal synaptic plasticity is claimed and only the 243 readout parameters learn.
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.