Looking at the activations, this neuron activates very strongly (values around 22,000+) on nearly every single token across multiple diverse documents - from book reviews to sports commentary to pest control ads to historical narratives.
The neuron appears to find **tokens in general text (
