There is no such myth. Nobody is surprised that you can say something about a distribution if you know some facts about the distribution. Using "single data-point" in two completely different ways is just bait and switch.
Any data pertaining to humans would be a single data-point in the true sense, which in itself gives almost zero information. You can't compare that to knowing global properties of the sample or the distribution, it's completely different things.
That is, now that you've admitted that a single datapoint does in fact have the potential for providing extreme information in the extreme case, then you should expect to find (some) information in the other cases... so that's what he does. As long as you assume its not a purely random distribution, which he does not.