The premise of "novelty" ML projects like finding Waldo, playing Mario, etc. has been that the application could be of greater use elsewhere.<p>Has this been demonstrated? I've heard Watson got really good at Jeopardy, but in the enterprise world many were unsatisfied with the results.<p>Is it still considered more efficient to focus on small ML projects and then apply it to a bigger problem? It seems like we have enough algorithms that we can start making valuable ML products and focusing less on novelty applications.