Predictive vs. Winning Models

I'm simply aestheticizing my peculiarity. The point is that a model that accurately predicts the world is much less capable of acting within it. It's not that a map isn't the territory and there's some important parameter it doesn't take into account. Like a GPS that doesn't account for street vibes, elevation gain, and so on. It's about something else. Let's imagine a predator and its prey. Does the predator have an ideal behavioral model for all the animals it hunts? Does it even exist at the very end? Absolutely not. That's absurd. A mosquito, for example, has no idea about the personality, work, or other aspects of the people it bites. There are only heat, CO2, and vision. The simplest algorithm possible to fit into a predator's design. On the other hand, do you need a good model of yourself? In general, much more than in the previous version. Knowing how much effort it has, what hurts where, what will definitely work, and what definitely won't. This is very important. So, civilization forces you to create detailed predictive models because its own model is so important to it. But it sabotages simple models that help people win, even if it benefits people from disseminating them. For example, Bitcoin. Initially, cryptocurrency had only one function: to subvert regulations: cross-border transfers, anonymity, tax evasion. States would only benefit from a time traveler who buys Bitcoin and holds it—its market capitalization would inflate slightly, and holding it instead of regular payments would demotivate others from using crypto as a means of payment. Clearly, one person won't change anything, but now there's an army of idiots who will destroy crypto projects by using them as an investment tool, not a payment method. All dreams of becoming a superpredictor are built around the fact that civilization motivates people with the hype surrounding three kopecks, and that they should be making predictions, not winning. Winning models have properties that generally follow from physics and computer science: And now my immortal homie, a.k.a. Kitten GPT, will read the epilogue for me. Features of winning models: Less input - ignores unnecessary inputs Low latency - speed > accuracy Loss asymmetry - critical outcome, not average, is important Discreteness - yes/no instead of probabilities Locality - no global consistency Instrumentality - usefulness > truth Embeddedness - inseparable from the action/body Cheapness - minimal computation and memory Opacity - poorly explained Non-scalability - breaks down when generalized Exploitation - seeks pressure points, not descriptions Incompleteness - "enough for action" Stakes - risk > probability Degradation - dies when formalized