A research infrastructure that could support improving the variance of results without significantly impacting the mean would be an enormous boon to the world.

I came to suggest something similar.

I often wonder how much this applies today. Academic science is currently heavily driven by funding of popular, low-risk research.

Also, without meaning any disrespect to Feynman or the advice, it's easy enough to say something like "disregard what others are doing" when you've won a Nobel. It's another thing to say this to people in just about any other inequivalent scenario, which is nearly everyone.

That is probably what is desirable for the field as a whole. But as an individual, not a good idea to shoot for high variance.

Anyone who wants high mean low variance career should become an accountant or a dentist: the payoff is vastly better, and it takes a comparable amount of intelligence.

This low variance attitude is why we can't have nice things. We have more "scientists" alive now than at any time in human history, and it is boring and useless with virtually no downstream results.

Part of the issue is that scientific fields are very punishing if you deliver negative results, thus the incentive to "shoot for the moon" is greatly lessened.