Earlier quoted context omitted.
No dude- you were right and he was wrong.
No, he's saying that if you have a few outliers to begin with you expect to get more normal-looking data later on, drowning your outliers in them. If you get 5 heads initially and then another 5 heads and 5 tails, it's still closer to the mean than when you began.
That can't be right, since the each point in the data is independent. In a 50/50 coin-flip game where you get 9 coins come up heads, the probability that the next coin flip results in tails is still 50%. What you need to understand is that probability deals with uncertain events. So instead of thinking about the final count needing to be around 50% heads and 50% tails, you should expect the count of the yet undecided part to converge around 50% heads, 50% tails.