First, I think we should seperate two ideas.

1. Creating a reference class.
2. Thinking in probabilities.

"Thinking in probabilities" is a consistent talking point for Musk - every interview where's he asked how he's able to do what he does, he mentions this.

Here's an example I found with a quick Google search:

Yeah, I think in general you always want to try to think about the future, try to predict the future. You’re going to generate some error between the series of steps you think will occur versus what actually does occur and you want to try to minimize the error. That’s a way that I think about it. And I also think about it in terms of probability streams. There’s a certain set of probabilities associated with certain outcomes and you want to make sure that you’re always the house. So things won’t always occur the way you think they’ll occur, but if you calculate it out correctly over a series of decisions you will come out significantly ahead…"

So that covers probability.

In terms of reference class, I think what Thiel and Musk are both saying is that previous startups are really bad to use as a reference class for new startups. I don't know if that means they generally reject the idea of reference classes, but it does give me pause in using them to figure out the chances of my company succeeding based on other similar companies.