>LLMs are simply manipulating symbols, they do not have semantic understanding.

falsify this. Show me a way you'd be able to prove they do/don't, that would work for humans.