abstract: Averaging theory allows to effectively describe the motion of a system with fast and slow components for a moderately long time. Yet, in applications it is often necessary to have information on the behaviour of the system for times much longer than the time for which averaging holds true. It is then natural to try to explore what happens for longer times. I will discuss some results pertaining the case in which the fast degrees of freedom perform a chaotic motion and will emphasise the similarities with the case in which the fast variables are described by a random process.