Intelligence is the manifold that these brute-force algorithms learn.
Of course we don’t brute-force this in our lifetime. Evolution encoded the coarse structure of the manifold over billions of years. And then encoded a hyper-compressed meta-learning algorithm into primates across millions of years.
Fitting a manifold to a bunch of samples does not allow you to understand what can happen in the universe. For example, if you train a regular diffusion model on correct sudokus, it will produce sudokus with errors because it does not understand the rules.
You raise a good point for the diffusion case, which trains only on positive examples, but generally speaking negative examples will warp the manifold appropriately.
Of course we don’t brute-force this in our lifetime. Evolution encoded the coarse structure of the manifold over billions of years. And then encoded a hyper-compressed meta-learning algorithm into primates across millions of years.