There are many ways to recall Paul Erdős' memory and his special way of doing mathematics. Ernst Straus described him as "the prince of problem solvers and the absolute monarch of problem posers". Indeed, those mathematicians who are old enough to have attended some of his lectures will remember that, after his talks, chairmen used to slightly depart from standard conduct, not asking if there were any questions but if there were any answers.In the address that he forwarded to Miklós Simonovits for Erdős' funeral, Claude Berge mentions a conversation he had with Paul in the gardens of the Luminy Campus, near Marseilles, in September 1995. After Paul's opening lecture for this symposium on Combinatorics, Berge asked him to specify his beauty criteria for a conjecture in discrete mathematics. Erdős mainly retained the following five:(i) The simplicity of the statement;(ii) The expected difficulty of the solution (which Paul liked to measure in dollars); (iii) The posterity of the subsequent theorem, i.e. the set of results arising either directly from the solution of from the methods designed to obtain it;(iv) The future of the path opened by the problem, which I would rather call the set of descendants of the problem, in other words the family of new questions opened up by the statement or the solution of the conjecture;(v) The intuitive representability of the specific mathematical property that is being dealt with. Apart, perhaps, the last, for which an adequate transposition should be described with further precision, these criteria are equally relevant to a classification for a conjecture in analytic and/or elementary number theory.My purpose here mainly consists in illustrating these criteria by revisiting some of the problems stated by Erdős in his profound article [24].Aside from updating the status of a number of interesting questions, my hope is to convince the reader that Erdős' conjectures, although stated in a condensed and seemingly particular form, were problematics rather than problems. Day after day, year after year, each of his questions appears, in the light of discussions and partial progress, as a node in a gigantic net, designed not for a single prey but for a whole species.In the sequel of this paper, quotes from the article [24] are set in italics. I took liberties to correct obvious typographic errors and to slightly modify some notations in order to fit with subsequent works. Erdős' paper starts with the following.First of all I state a very old conjecture of mine: the density of integers n which have two divisors d 1 and d 2 satisfying d 1 < d 2 < 2d 1 is 1. I proved long ago [20] that the density of * We include here some corrections with respect to the published version.1. We shall make use of this extra information later on. 2. See [29] for a short proof.