abstract: The adaptation of a biological population to a new environment can be conceptualized as a random walk in genotype space biased towards increasing fitness. The accessibility and repeatability of such evolutionary pathways is governed by an interplay of the fitness landscape topography and the distribution of fitness values, both of which are difficult to obtain from first principles but can, to some extent, be inferred empirically. In view of the scarcity of experimental data on fitness distributions, it has been suggested to make use of extreme value theory to classify the possible asymptotic shapes, from which the expected evolutionary dynamics could then be predicted. The talk will review these developments, starting from the repeatability of a single adaptive step and subsequently focusing on recent results for adaptive walks on correlated fitness landscapes.