Polysemantic neurons are a major challenge for the circuits agenda, because N meanings in one neuron times M in another creates NxM effective connections that cannot be considered individually.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• We hypothesize that polysemantic neurons may be resolvable by unfolding networks or training to avoid polysemanticity.hypothesis0.829
  Forward-looking proposal for how the polysemanticity challenge to circuits research might be overcome
• There is a many-to-many mapping between neurons and concepts, meaning multiple high-level causal variables might be encoded in overlapping groups of neuronsclaim0.804
  Fundamental theoretical claim motivating DAS, attributed to Smolensky/Rumelhart/McClelland.
• No established method for resolving polysemantic neurons into pure features at scalequestion0.800
  Identified gap linking polysemanticity challenge to disentangled representations literature
• Polysemantic Neuronconcept0.798
  A neuron that responds to multiple unrelated inputs, posing a major challenge for circuit-level interpretation
• Models with 1-hot activation sparsity still have polysemantic neurons; single neuron trained on 4 mutually exclusive features prefers polysemantic representation with loss ~0.7 vs 0.8finding0.792
  Counter-example disproving that architectural sparsity alone can prevent polysemanticity
• Superposition is in some sense deliberate: the model converts pure neurons into polysemantic neurons to store more features in fewer neurons.claim0.780
  Interpretation of the cars-in-superposition circuit finding as an intentional representational strategy
• The huge chemical complexity within each synapse suggests that neural models of cognition that refer only to changing weights of synaptic connections and ignore sub-neural chemistry are probably ignoring some of the most important explanatory mechanisms in brains.claim0.777
  Sloman's critique of mainstream neural network theories.
• Neurons can correspond to interpretable functional roles but interpretations in terms of individual neurons are unlikely to be the most parsimoniousclaim0.774
  Claim from footnote 3, acknowledging neuron-level interpretability while arguing subcomponents are better.