The point of a competition is to meet specific parameters as well as possible and push the boundaries of what can be done. It's not meant to create a "daily driver".
I realize he argued against this with the coin flip test, but that is why you'd ideally want to have many of these competitions over time. If you start to see the same names popping up at the top regularly, you know there's some sort of significance to them. Teams would ultimately want to trend towards whatever wins competitions most consistently, so they'd want to rely on models they think are the most likely to perform in a real world test. They wouldn't want to simply rely on a coin flip.
And in large competitions, you have the chance of batching the top performers together and seeing what is common between them. Presumably there's a reason these models pan out over the rest in aggregate; are they worth pursuing a bit?
I think we agree at the end though, even within my analogy. A huge value of F1 racing is the publicity the teams give their sponsors. They might learn some information that can be pushed down to their consumer vehicles, but it's marginal compared to team winnings and the value of saying "See? Our engineers are the best".