~/polysemanticity

25 views

member of (1)

containment 1 · border 0vote §

~/polysemanticity

prompt

A neuron that appears to have several unrelated jobs, or no legible job at all. Rank the explanations by how much of observed polysemanticity in real networks each one actually accounts for, after you have looked hard. Source for the candidate list: Cammarata, "Neural Networks Are Already Clean" (Goodfire, 2026), section "Superposition of the Gaps".

discussed in #polysemanticity

ranked child groups
scope depth
vote on children

no voted pairs yet in this scope

score history draws itself once two or more members carry votes — vote on children

unranked

~/arbitrary-packing
~/bad-learning
~/bad-microscope
~/recursive-generalization
~/unread-pattern

cli

src

spread

↑ member of (1)

↓ ranked child groups scope depth 123∞ copy vote on children

:share-of-polysemanticity vote

member of (1)

ranked child groups
scope depth
vote on children