[Patrick Sileo]

Game theoretic equilibria have the property that no individual player can change his action and make himself strictly better off (vs the equilibrium strategies of the other players). They come in two forms: pure strategies and mixed strategies. The former consist of a set of non-random actions (one per player). If a player takes an out-of-equilibrium action, he may be worse off. Since poker is a zero-sum game, when your opponent is worse off, you are (generally) better off.

In a mixed-strategy equilibrium, each player is randomizing over a set of non-random actions. Let these actions be say {A,B,C}, selected with probabilities p(A), p(B), p(C), respectively, where p(A)+p(B)+p(C)=1. It is a property of this type of equilibrium that a player's expected payoff is unchanged if he chooses a different randomization over the same strategy set {A,B,C}. In this sense, mixed-strategy equilibria are "mistake-proof". However, should an "incorrect" action "D" be included in the randomization, the player now may well be worse off, as in the pure strategy equilibrium case.

In general, game theory's non-exploitable strategies are most useful against better opponents, since you cannot be exploited. On the other hand, it can be worth playing a theoretically exploitable strategy against an inferior opponent so that you can fully exploit him.