paulfchristiano comments on Concept Safety: Producing similar AI-human concept spaces