Videos have irremovable noise, but in some domains there is none and compression is more useful. In my experience, one example where the prediction-compression duality made problems much easier for humans to understand is in code golf (writing the shortest possible programs that perform various tasks). There’s a Stack Exchange site dedicated to it, and over the years people have created better and better code-golf languages. Solution sizes shrunk by a factor of >3. Now they’re so good that code golf isn’t as fun anymore because the tasks are too easy; anyone fluent in a modern golfing language can decompose the average code golf problem into ten or twenty atomic functions.
Videos have irremovable noise, but in some domains there is none and compression is more useful. In my experience, one example where the prediction-compression duality made problems much easier for humans to understand is in code golf (writing the shortest possible programs that perform various tasks). There’s a Stack Exchange site dedicated to it, and over the years people have created better and better code-golf languages. Solution sizes shrunk by a factor of >3. Now they’re so good that code golf isn’t as fun anymore because the tasks are too easy; anyone fluent in a modern golfing language can decompose the average code golf problem into ten or twenty atomic functions.