Most interesting to me was the claim that around a fifth of the system needs to be in exploration to prevent monoculture collapse. Their graph showed that at 30% you had 100% surviving diversity. Where exactly where did this data come from? What was the methodology? Why is it so linear? I'm most curious because if it's a robust figure, it should translate to the myriad of systems we have plagued by diversity collapse.
Other than that, a lot of good data points under one roof looking at our monoculture problem across verticals.
I kind of feel bad sharing this from my lecture backlog because it seems every professional YouTube creator is suddenly talking about how AI and compression are related. However, here's a video not in that sphere of influence if you, like me, just really want to know more about the nerdy details involved in compression applications in real, well, applications. Things like using UDP packet lengths as a way of encoding extra zero bytes between packets.
If you're interested in the state of the art, I know Fabrice Bellard released v3.3 of NNCP not too long ago. Just know, it's probably not a good daily driver. It's designed for maximum compression over English text with less regard for the memory and time required. Again, compression is all an intelligence game.
