We study the geographical patterns of scientific collaboration from a large sample of research papers and letters written by two authors that appeared in the magazine Nature over two sub-periods, before and after the popularization of Internet use. We report three results: First, the distance distribution of coauthors is fat-tailed, in agreement with other studies that find a gravitational law in collaboration networks. Second, in the later period the distance distribution dominates the range of commute-distance and beyond (>50km), which renders the city the atomic unit for statistical testing. Last, strong geographical clustering remains a major generative factor in this network. Assuming the universality of this law, we estimate the gravitational constant from the pull between scientists in the network. We find that this constant has decreased two-fold over the last three decades while the other coefficients remain stable. This may indicate that the gravitational constant absorbs changes in the environment that render distances easier to cross, namely a "lighter world"