vintage screwlisp account on Nostr: the idea that occurs to me is that a probabilistic decision tree, even generated to ...

the idea that occurs to me is that a probabilistic decision tree, even generated to fit some data, is exactly a probabilistic decision tree, even a non-trivial one (halting problem one). Whereas an attempt to universally approximate from sampling the same data has a universe of possibilities, probably exacerbated by having a pre-trained model for example. I'm not sure how to formalize that idea though, and I have many pages on the topic to read.