This week Bill reflects on the progress made using transformer networks to integrate language and visual processing with robotic control, particularly in Google DeepMinds RT-2, and whether we're heading for embodied awareness, while Gareth speaks out for the special place well-crafted audio has in the cultural landscape - who needs pictures? And we go purple about the Vesuvius Challenge. (Image of cocktail making robots taken at the Barbican 'AI: More than Human' exhibition in 2019. The Negroni was acceptable.)