Evolution of random systems as well as dynamical systems with chaotic (stochastic) behavior traditionally (and seemingly naturally) is described by studying only asymptotic in time (when time tends to infinity) their properties. The corresponding results are formulated in the form of various limit theorems (CLT, large deviations, etc). Likewise basically all the main notions (entropy, Lyapunov exponents, etc) involve either taking limit when time goes to infinity or averaging over an infinite time interval. Recently a series of results was obtained demonstrating that finite time predictions for such systems are possible. So far the results are on the intersection of dynamical systems, probability and combinatorics. However, this area suggests some new analytical, statistical and geometric problems to name a few, as well as opens up possibility to obtain new types of results in various applications. I will describe the results on (extremely) simple examples which will make this talk quite accessible.