Here we formulate the Connes embedding problem, whether any tracial probability space embeds in an ultraproduct of matricial ones. We also briefly describe the so-called hyperfinite factor R, with which one can reformulate the question as asking for an embedding in an ultrapower of R. Since the Connes problem is open even for the tracial probability spaces associated to discrete groups, this leads us to describe several related interesting classes of infinite groupssuch as residually finite, hyperlinear and sofic groups. We also discuss the so-called matrix models in terms of which the Connes problem can be naturally reformulated. Lastly, we give a quite transparent characterization of nuclear von Neumann algebras, which shows that there are very few of them.