No, there is research in that direction and it shows some promise but that’s not what’s happening here.

Are you sure? It would be great to get official/semi-official validation that thinking is or is not resolved to a token embedding value in the context.