Benjamin Van Roy: TD(0) Leads to Better Policies than Approximate Value Iteration. NIPS 2005: 1377-1384