Doing old things better is very amenable to metrics, lifetime value of customer, market-adoption forecasts, cost of customer acquisition, etc. This is the land of venture money with milestones and ever-growing rounds as success unfolds. (Or asphyxiation if success doesn't happen.)
Doing brand new things probably is far more likely to start via bootstrapping, passion projects, university research and other areas where open-ended curious inquiry is tolerated better. Dixon (OP) loosely acknowledges this when he writes that entrepreneurs "usually spend many years deeply immersed in the underlying technology before they have their key insights."
Eventually the path forward will be clear enough that the VC/metrics folks can show up for later financing rounds and do their thing. But I'll argue that venture's metrics-driven feed-or-starve system is more likely to mess up this early exploration than to help it.
Even though VCs will insist that they are equally good at both.