Why shouldn't they have access to the American version of the next token predictor?