Well, sorta. There is some evidence to suggest that neural networks learn 'universal' features (cf Anthropic's circuits thread). But I'll openly admit to being

Well, sorta. There is some evidence to suggest that neural networks learn 'universal' features (cf Anthropic's circuits thread). But I'll openly admit to being out of my depth here, and maybe I just don't understand OPs point