Do you think we'd need to cross a certain size threshold of the network (>8b, >70b, >300B, >700B, ...) for a multimodal model, for this phenomenon to be worthwhile investigating? Because unless Wanting is an observable phenomenon in small networks, the model sizes used in current studies about multimodal representation wouldn't suffice. And I'm not aware of any small (~8b) networks showing the "I'm pulled towards" behaviors that Opus shows.