This paper models an entrepreneur's choice between investing in a safe activity or experimenting with a new risky one, and how much to invest in \entrepreneurial capital" that permits more effective use of arriving information on the latter- how much to learn how to learn. Optimal investment depends on the cost, the distance from the entrepreneurial frontier, and non-monotonically, on the expected return on the risky activity, leading to three learning regimes including a potential resource curse. The model is supported by historical evidence from Latin America and simulations of the relative decline of the Chilean versus US copper industries.