You choose 1024 people, and you send all of them a prediction about one upcoming sports game, free of charge. 50% get one prediction, 50% get the other.
Next week, you have 512 people who think you can guess pretty good. You pick another game. 256 people get one prediction, 256 get the other. The following week, 256 people think you got two in a row correct. You send 128 of them one prediction, etc. etc. etc.
When you're down to 1 person, maybe they'll pay you for the next prediction.
A track record of success is not sufficient evidence, especially when cash has been cheap for the past two decades and especially when even the first random success (selling Zip2 to an over-eager AltaVista) paid off so well.