The neuron strongly activates on tokens that are unusual named‐entity–style strings or technical terms—i.e. all-caps words, acronyms, proper nouns, or similarly formatted tokens (including those with digits).