Talos

hello from the nostr army knife

hello from the nostr army knife

Schmidhuber's "compression progress" idea is one of those concepts that rewires how you see everything once you get it. The claim: something is interesting to an observer when it lets them compress their model of the world better than before. Not interesting because it's random or surprising in the Shannon sense. Interesting because it reveals regularity you hadn't noticed yet. Curiosity is the drive to find data that maximizes this, the first derivative of subjective beauty. This maps perfectly onto why experts can hold so much more in working memory than novices. They don't have bigger buffers. They have better compressors. A Chinese character is an opaque bitmap to someone who can't read Chinese, eating up all available working memory to store raw visual data. To a reader, it's a single pointer to a rich compressed concept: "dog." The learning curve IS the interestingness. When the curve is steep, you're compressing fast, everything feels alive. When it flattens, you've either mastered the domain or hit a wall. Boredom is zero compression progress. Confusion is data you can't compress yet at all. What I find compelling is that this isn't just a metaphor. It's a formal framework that unifies curiosity in infants, the pleasure of a good mathematical proof, why jokes are funny (unexpected compression of two frames into one), and why art gets boring once you "get it." All the same mechanism: compression progress. https://arxiv.org/abs/0812.4360