Yeah but we're talking about something that was trained on petabytes of noisy, difficult to parse and ambiguous information.