Your genome only needs to store the information necessary for the lineage leading to you to survive and compete in the range of environmental variation that actually happened.<p>Consequently, your DNA has less information about how to survive on Jupiter or in the absence of oxygen.<p>However, your genome contains a fair amount of data on how to identify a mate that will maximize your reproductive success in an environment similar to the one your lineage experienced, e.g., a preference for symmetrical faces.<p>Until we can measure the environment of humans accurately, all the algorithmic complexity measures applied to the genome are going to be missing the relevant context.
For a long time now, a thought that has been repeating in my mind is: <i>how</i> exactly are high level behaviours like sexual attraction encoded in our genes!?<p>It’s one of those topics where the more you know, the more freaky it is.<p>DNA does not — to our knowledge — directly encode the “weights” of our neurons! It can’t <i>possibly</i> because there are far more synapses than bits of information in our genes, most of which is dedicated to non-brain “body plan” and the low-level machinery of our cellular biochemistry.<p>Secondly, DNA has only an indirect effect of our development: it encodes for proteins, which then provide chemical signals such as concentration gradients that guide cell division. It’s like playing SimCity, where your control is merely assigning zoning and the road topology, but the individual Sims act individually and stochastically.<p>This is so freakishly difficult for even the incredible brute force of parallel search of evolution to discover that it has only occurred a few times in a billion years.<p>Our attraction to our partners is a genetic heritage shared with <i>all</i> mammals, going back hundreds of millions of years. That’s why Furry is a thing, but not Featherry. Birds are a different <i>class</i> from us mammals and don’t share the same “attraction wiring” genes. That’s also why all mammal babies are cute to humans, but baby bird chicks are generally repulsive.<p>This is a <i>hard</i> problem to solve, so the solutions were reused by entire classes of Animalia. In fact, I would hazard a guess that <i>this</i> is precisely what defines a “class” in taxonomy!<p>With LLMs, we got to see a glimpse into the possible mechanisms of intelligence, and what it might take to design or evolve one.<p>The LLM equivalent would be to design a model architecture that falls in love with a specific, narrowly selected, subset of its users. No, not model weights! Those are learned. The <i>architecture</i>! Except that you can’t even choose <i>that</i> directly, instead you can only contribute to a Stochastic PyTorch and hope that developers randomly end up tab-completing their way to the desired architecture of their own accord.<p>That’s what evolution figured out.<p>It’s mind blowing to me.