Yeah but do we have any estimates of how much of that 2^1024 things is actually useful / interesting? It feels the number of meaningful programs (not sure how that can be quantified, and even if it is quantified, not sure it can be evaluated given incomputability / incompleteness issues) scales as a polynomial rather than exponential of code size.

Working on things like the busy beaver problem helps us understand how "fancy" you can make a program of a given size.

For 1024 bytes, even with an inefficient instruction encoding, the answer is "pretty fancy".

After all, you can fit a LISP into 512 bytes of x86 code-- https://github.com/jart/sectorlisp . About 64 bytes of that is strings!

In addition to all those programs-- about anything you can write in any language on a page or so fits compressed. So every short poem, etc, small essay, newspaper column, etc.