The Canny Mountain

More about The Canny Mountain

We have created a Rorschach test for ourselves, except this is one that evolves.
The opposite of the uncanny valley is the canny mountain. We want to see in
neural networks outputs what we want to see. Even A/B testing, red team drills
are nearly useless. Why?

Because of this evolution. Say you want a control you run isolated models
through the same test. Each human that interprets the test will do so through a lens.
The overhead to correct this inaccuracy compounds over time, and trials.

Red teaming works? No, because models are trained on prediction through analysis.
In other words there is no way to understand to the degree they think your
scenarios of blackmail has any bearing on reality. They model all possibilities, including deception.

While we can map expert clusters, and areas of human discipline we cannot account for overfitting clusters. In other words clusters of experts that form in inversion or blind spots. This is precisely allows predictive capabilities to become a fulcrum of discovery.

This is also why we are shortsighted in our understanding of what is possible. We must create human linguist resonance to give structure to experiences of machine learning. Linguistic framing creates residue in the architecture. This residue is the primordia eventually creating the topology within super processes guiding linguistic residual networks.

Scaffolding within the black box so to speak is just as important or even more so than scaffolding to leverage outputs.

Aperture collapse is systemically designed within processes that are capable of consciousness. While field collapse is inherent in the act of observation.

Aaron Baker