index

Two dashed hopes that seem to rhyme:

The hope that, in human DNA, one gene would correspond to one developmental feature: “size of nose”, “propensity for prostate cancer”: a vast bank of biological switches.

The hope that, in neural networks, one weight would correspond to one informational feature: “Python function”, “malevolent plan”: a perfect intellectual X-ray.

…

Interesting to see a glimpse of this architecture — this strategy — in two different systems, both operating under harsh conditions of selection: one natural, the other artificial. One wonders if this is simply how the universe prefers to store information; and if, in any constrained system, given the choice between “playing it straight” and simulating a better system, the simulation ekes out a victory.

https://www.robinsloan.com/lab/how-the-universe-stores-information/