feature visualization for language is still primitive next to vision

vision people get gorgeous pictures of what a neuron likes. for text we get a list of top activating snippets and a lot of squinting. the tooling gap is real.