mental handles I use a lot but haven’t translated for casual communication:
- People’s expert intuition, generalized as them having an subnetwork that has fine-trained on a particular kind of data.
- Approximating real things as simple math/logic, with a modelable AMOUNT or KIND of divergence from the real values. I visualize this as the Area between two function curves.
- The idea that many models will behave functionally identically over restricted DOMAIN, and only diverge in the extrema or in less common cases. I may not know what the real function is, but I can point at the tight convergence of predictions as a pseudo-function that may or may not have a continuous extension outside of that domain.
- similarly, the idea of multiple lossy models covering different domains, which pieced together suggest there could be a continuous simple function that works well over the combined domain.
- the skill of blurring your mind’s verbal/symbolic/social reasoning, to just look at what strictly literally happens.
- pointing at the most comfortable subcognitive compressions for complex data in the human brain. Metaphors that we keep settling on like ENERGY FLOWS and SOULS and LUCK, that unfold fractally into a bunch of tiny correlated details. Powerful, but not necessarily as a predictive model.