1. Along the way, Brad DeLong mentions Milton Friedman's famous claim that a model is better the more unrealistic its assumptions, and that the sole measure of a theory is making accurate predictions. I'd really like to know what DeLong thinks on this, but his views aren't there in the interview. He mentions Friedman's idea but doesn't defend or attack it, just a reference to one of the most influential ideas on this topic, I guess. Shows how much Friedman's view is still in play.
In my view (some not very well organized thoughts here) the core problem with Friedman's argument is that a theory with perfect predictions and perfectly unrealistic assumptions simply doesn't teach you anything -- you're left just as mystified by how the model can possibly work (give the right predictions) as you were with the original phenomena you set out to explain. It's like a miracle. Such a model might of course be valuable as a starting point, and in stimulating the invention of further models with more realistic assumptions which then -- if they give the same predictions -- may indeed teach you something about how certain kinds of interactions, behaviours, etc (in the assumptions) can lead to observed consequences.
But then -- it's the models with the more realistic assumptions that are superior. (It's worth remembering that Friedman liked to say provocative things even if he didn't quite believe them.)
2. An interesting quote from economist James Galbraith, with which I couldn't agree more:
Modeling is not the end-all and the be-all of economics... The notion that the qualities of an economist should be defined by the modeling style that they adopt [is a disaster]. There is a group of people who say that if you're not doing dynamic stochastic general equilibrium modeling then you're not really a modern economist... that's a preposterous position which is going to lead to the reduction of economics to the equivalent of a small religious cult working on issues of interest to no one else in the world.