[jkingsbery]

I had a math professor in college, whom I had for Real Analysis. He told us that one thing that helped him learn was to ask somewhat-obvious sounding questions to try to make connections to things to check understanding. There are at least two good reasons for this: (1) if you don't understand the basic (unsurprising) things, you probably won't understand the more nuanced things, and (2) what counts as surprising varies with the audience.

If someone were to come to me looking for advice along these lines, I'd say: sure, focus on the surprising thing, but it has to be grounded in the familiar, and what counts as "familiar" depends on the audience.

[massysett]

Precisely. I think the author is saying a talk should have a thesis statement that is novel. That is true.

But to convince the audience of the thesis statement, the speaker needs to cite evidence. The evidence cannot be surprising.

Since most of the talk needs to be evidence, and therefore not surprising, it’s foolhardy to eliminate all that is not surprising.