302 neurons.
One search for food.
Observe, train, and change its neural network. Explore what this worm model can learn.
Biological connectivity
Simplified behavior model
Tokenized stock holdings
Recent token inflows Tax source unverified
The NEURO token CA is listed above. Tax attribution is not yet verified. Inflows cover recent ERC-20 transfers of USDT, USDC and listed tokenized stocks, excluding BNB transfers. Inflows are not trading profit. Signing is disabled. Holdings show on-chain token units, not share counts or valuations.
How does this experiment work?
Chemical synapses and electrical connections between 302 neurons are extracted from the Cook 2019 hermaphrodite connectome, preserving recorded directions and weights. Positions are illustrative, not anatomical. Each step runs four simplified tanh propagation rounds from sensory inputs. All weights are positive; biological excitation, inhibition, temporal dynamics and muscles are not reproduced.
Food direction and adjacent obstacles are manually mapped to ASEL, ASER, AWCL, AWCR, ASHL, ASHR, ALML and ALMR. This encoding does not represent biological cell functions. The 302 activity values feed a trainable four-action linear readout. Shortest-path demonstrations and cross-entropy gradients update only the readout, leaving connectivity unchanged. The body animation interpolates the chosen actions.
Test maps use fixed seeds separate from training and allow up to 160 steps. Disabling a neuron sets its activity to zero, affecting propagation and actions. Results describe this simplified model on this task, not biological learning or consciousness. Refreshing clears training; retrain to reproduce the experiment.