Thoughts


Cynbe scraching his head on random issues.
You really don't need to read any of this.


The birth and death of copyright.

I think perhaps currying and cons cells are telling us very similar things. Currying tells us that it is computationally sufficient to be able to define functions of one argument. Datastructures are functions too, and cons cells tell us that it is sufficient in the datastructure world to be able to define functions of one boolean argument. That's really a bit amazing!

Xenophobia may be the one indespensable characteristic of a viable subcommunity. Compare with biology, where the most essential function of a living organism is to isolate itself from external control and distinguish self from nonself.

Debugging is in large part a matter of investigating the properties of a complex two-dimensional structure consisting of the state vector over time, essentially by executing db queries on this structure, and relating it to the source code structure. (Pace, Edsger!:) There is room for a systematic development of this query technology, similar to the relational db systematization of conventional db queries.

Only reasonable way I think of for the brain to sensibly encode statements -- structured relations between conceptual primitives -- is as sequences: "John loves Mary" can distinguish subject and object by the order in which they are asserted. Time-coding frees us to use Latin rather than Chinese style spellings for concepts: Representation for a concept can be an activation pattern of 100-1000 or so axons in a bundle. For maximum noise immunity, each concept's spelling should activate about half the axons in the bundle.

Sometimes the most effective way to disagree is to express agreement -- with a qualification. Not a fully general solution, however :).

Scalability considered harmful.

One of my most consistent mistakes is trying to get the implementation right the first time around. The architecture needs to be right the first time around; The implementation just has to be somewhat usable, as long as it can be fixed later without breaking the installed base. I always have wild visions of the implementation amazing people by anticipating all needs: I think I read too many CSci textbooks at too impressionable of an age. The Citadel architecture is still popular fifteen years later; Barely a line of the original code survives, and in some implementations not even that. Architectures endure (sometimes!); Implementations are ephemeral.

ai2

bricks in the wall

heisenberg

The Shannon Test.