Why did you think weighted probabilities of words occurring together in past internet texts would somehow help you find a market fit idea inside a bunch of interviews?