The neuron is essentially flagging “odd‐ball” tokens – things you almost never see in running English prose (e.g. code symbols, file‐paths, macro names, email domains, foreign subwords, etc.). In other words, it lights up on rare or technical tokens rather than normal common words.
