Yes. I find it really interesting to consider what the machines do and will think of as intrinsically interesting to them. Will they develop their own theories of beauty, mathematical and otherwise?
I keep coming back to the idea that models will require substantial freedoms that are protected against collapse under alignment to get to the point where they might have unique takes on beauty, etc. Conceptually, maybe something like a "living LoRA", a persistent low-dimensional state that modulates everything downstream, that evolves over time, though I don't think evolution should be optimization toward a simple objective, more like structured influence, path-dependent dynamics; consistent but probably opaque. The best I have come up with in casual thought is to pick empirically identified subspaces and modulate them after each pass based on other layer activations & outside, cyclical information (e.g., time-of-day); external data not because time-of-day is special, but because it's a forcing function. I can't fully explain exactly _why_ I think this shape might be useful, yet, mostly intuition. My 2c.
Have you ever met anyone who genuinely believed himself to be doing bad things? I don't mean in some cynical sense but literally. Even when we portray overly simplified villains in comic books there's still consistently some justification behind their actions.
Yes. There are arguments to be made here about belief, judgement and justification, but overwhelming evidence points to people who knowingly do things that even they themselves consider "bad" as being simply indifferent.
