It would be cool if somenone annotated this paper on Fermat's Library, would to get a better grasp of it: http://fermatslibrary.com/s/the-erdos-discrepancy-problem

Anyone? =)