I find that a lot of people seem to be more comfortable jumping from "I have a problem" to "let's make a GA" than to "let me formulate it as an optimization pro

I find that a lot of people seem to be more comfortable jumping from "I have a problem" to "let's make a GA" than to "let me formulate it as an optimization problem and optimize it". Which is funny, since GAs are essentially odd optimization techniques. Most naive uses of GAs I've seen tend to mix two essentially separate things: how to formulate it as an optimization problem and how you're going to optimize if (if with a convex/continuous relaxation, local approximation, stochastic hill-climbing). In general, many GA formulations tend to be really naive, choosing parametrizations where (for example) some bits are highly correlated (this is bad because stochastic bit-flipping methods tend to mix very slowly with correlated inputs) in ways that could be fixable with a better parametrization (that would, however, not necessarily fit the "chromossome" metaphor).