Why not both? Berkeley AI Blog, Stanford AI Blog, CMU ML blog... all show that you can do both. Reviews (part of paper submission process) are legitimately usef

Why not both? Berkeley AI Blog, Stanford AI Blog, CMU ML blog... all show that you can do both. Reviews (part of paper submission process) are legitimately useful, if done well. As a researcher, having Arxiv as a standard medium and conferences to help filter interesting papers (if imperfectly) is also useful.