In the universal graduate-student tradition of reading papers to procrastinate, and because it was well received last time, I've collated a handful of fun articles from the arXiv from the past few weeks.
Terry Tao has (probably) solved the Erdos Discrepancy Problem. Of particular note is that some of the methods used in his proof grew out of the massively collaborative polymath project spearheaded by Tim Gowers. The Discrepancy Problem is one of those easy-to-understand, outrageously-difficult-to-prove problems in number theory: it involves whether or not partial sums of infinite sequences only containing {+1,-1} necessarily grow past a certain bound. Paul Erdos offered a $500 reward for solving it, back when it was originally posed. See Tim Gower's blog for a more technical writeup.
"Loophole free" Bell Inequality violations. These sorts of experiments have a long history. Suppose you're playing a game with some dice. Bell's theorem says that if you have a pair of suitably defined "quantum dice", you can win a particular game more often than you could ever win that same game with a pair of boring old classical dice. This simple result ends up having fairly huge implications for the structure of reality--big enough implications that people really, really, really didn't want to believe that it was possible to do better in this game with quantum dice. "Loopholes" are the various ways in which one might think one has "done better", or in the parlance of the field, "violated Bell's Inequality", but in reality, one has cheated somehow by way of a "loophole". Suffice to say, if quantum mechanics doesn't bother you, then Bell Inequality violations also don't really bother you. But quantum mechanics bothers some people a lot, and this experiment is a final nail in the coffin for people that doubted the philosophical ramifications of quantum mechanics.
Google's deep-learning scheme to learn to play videogames is now better than human players in 31 games. This is up from the 23 games a few months ago. There's a joke here about machine overlords, but it's sort of stale, and also no longer a joke (see also, a different AI does as well as a 4-year-old on an IQ test).
Why now is a great time to study Quantum Computing. This is actually from nine months ago and actually just a shameless plug for my field.
Observation of pentaquarks at CERN. 15 sigma. 15! Pentaquarks are interesting analogous to the way a hard-to-synthesize molecule is interesting. They're perfectly consistent with known theory, which makes them less interesting than something like the Higgs Boson, but they're hard enough to probe theoretically that having an observation of them in the lab furthers our understanding of quark interactions. If your next question is, "Why should I care about quark interactions?" there's probably nothing I can say to convince you that they're interesting, short of an appeal to the intrinsic merit of fundamental science research (and then I might point at your cell phone and yell about transistors and quantum mechanics).
