Current AI models lack basic physical intuition like understanding gravity or how objects break, leading to mistakes on tasks requiring common sense. The fix: use them only where they fit and always have humans review the results.
One of the criticisms of this system right now is that it's currently— the current implementation of it is really around this meaning-type concept, but it still doesn't quite have what we consider common sense, which is basically we have built into our minds and our abilities the ability to model the world around us. We have a fundamental concept of gravity and that stuff's going to fall and stuff's going to break if you do certain things. Which this model, the model that's currently implemented doesn't really quite have some of that. And that's some of the criticism. And so it will make mistakes. And so you want to make sure you're using it for the right kind of things and that you have a review process.
