This neuron isn’t picking out any coherent semantic category in context; it simply responds based on each token’s learned bias (i.e. it activates whenever its associated word embedding has a large bias term), not because of any meaningful pattern in the text.
