On cloning games: It's a good eval for frontier models, since they still fail at getting the small details that make games great. Games are also very detail-dense, so models that fail to carefully inspect video will miss key mechanics. It is not meant to replace or degrade the original game. On vibe coding games: To get a picture of your family, you used to have to pay a painter to carefully paint you on a canvas while you all stood still. Then photography replaced that, and with digital photography and smartphones, anybody can immediately have a high-quality picture of anything. That doesn't mean there's not professional photographers; it's just not needed to get a picture. I think we'll see a similar trend in video games. Back to cloning: Using AI to clone a game is different than just taking a picture of an artwork because, right now, it's used as an evaluation. It doesn't produce a perfect replication, and it's useful for finding the differences between the best models and the best of the best models. If and when it's easy to instantly and perfectly replicate the game, there will be no further point in the test. Cloning a game will then be like screenshotting someone's artwork to hang on the wall. But we're not there yet. And when I clone a game, it's not so I can play it instead of the original. (I'm not a gamer anyway.)