I like some of this. That said, if the system takes previous states containing information about a car’s location and motion, transforms that information through a giant learned mathematical function, and produces a later state with the car appropriately displaced, something computational happened, no? It wasn’t necessarily a conventional physics simulation integrating velocity through every intermediate timestep. That’s somewhat fascinating. But “therefore nothing was computed,” I mean, not really.

Learning from stats (AI) has two big problems: it's non deterministic and much more expensive to compute than the deterministic alternative (when available).

The strength is dealing with ill-defined problems.

This last point is what makes it feel like magic, but the first two are the ones that can lead to disillusion.

The statistical approach only works if we can get computation cheap enough, sufficient data and if the problem can tolerate "good enough" results.

Thank you for taking the time to read my long essay and give your feedback! I think there are valid points here, and I will incorporate them to my revised edition soon. I'll reply some comments here individually.

>What does that mean? It means you can walk forward in a generated world, look at Stonehenge, turn around, walk away for sixty seconds, turn back, and Stonehenge is still there. Structurally intact. In the correct position. This might sound unremarkable until you consider that nothing in the system is tracking Stonehenge. There is no variable called `stonehenge_position`. There is no entity in a database. The model simply generates the next frame based on everything it has seen so far, and because it has learned how persistent worlds work, the monument persists.

Am I the only ignoramus who did not know the above? Until now I indeed thought that the lack of persistence in the world was indeed "fixed" in newer video models by some sort of "remember the objects you just displayed"/"remember the objects behind the camera" code/storage.

>Baudrillard had a word for this, back when the most advanced simulation available was Disneyland. A simulacrum: a copy with no original. The 2 in that cell is not a copy of a computation. There was never a computation to copy. It is a symbol that refers to nothing except what a 2 usually looks like in that position, and it happens to be right.

We've all heard the saying "fake it 'til you make it".

We're now approaching the point where the "fake" is more real than "reality".

I really enjoyed the writing in this.