Thanks for the kind words.

We've had the same thought, and are experimenting in this direction in the context of recursive language models.

Let us know if you have feedback!