science
AI picture naming is not consistently localised to one attention head
A bioRxiv preprint tested picture naming in six vision-language models by silencing each attention head in turn. The results ranged from one strongly necessary early head to no critical head, showing that disruption and computation are not the same.