The *ideal* video compression algorithm, fed random bytes to decompress, will produce a normal movie with a coherent plot. Contemplating this may help make clear why “compression is intelligence”. It’s also literally how GPT-3 works.

I clarify: that was truth, not humor. The GPT setup is precisely isomorphic to training a (huge) neural net to compress online text to ever-smaller strings, then using the trained net to decompress random bytes.

And to double clarify, by “precisely isomorphic” I don’t mean “vaguely analogous”, I mean that if you give me the trained GPT net, I can use it directly as a text compressor and decompressor, with no retraining; that’s just what it already is.