Self Improving AI is getting wild

Self improving my bank balance is what I need.

Something that many evolution algos miss is the concept of sexual selection - That an evolution algo isn't just about single-generation environmental fitness, it's also about the ability to detect future fitness. The sexual selection mechanic represents the ability of the organism itself to select successful versions of itself without requiring environmental selection to take place. It provides an alternative selection mechanism, one which selects over the longer term compared to a single generation selection mechanic. In a single generation you may be extremely unfit, but environments change, and fitness over multiple generations is more important than single generations. What is alive today is what survived the longest, not simply what was fittest.

1:35 This drawing is even better when you learn that Jurgen drew it himself hahaha

C for concept of agnosticism

I think generally the last author is the team lead.

Saying WordPress is a good choice because half of the world's websites use it, is like saying poverty is a good choice because half the word lives in poverty.

Wes - the image, I think, is a motif to the Baron Münchhausen story which gives rise to the Münchhausen trilemma: landing on a system that lifts itself by improving the very process it uses to improve.

Try Vibe for WordPress here: https://10web.io/?utm_source=Youtube&utm_medium=Influencer&utm_campaign=WesRoth

Excellent explanation! When I scanned the schmidhuber paper a couple of days ago I didn't realize that it was a 'competitor' for the Sakana AI paper. Thanks.

Love your videos Wes, thank you. This video really got me thinking, specifically how the archive tree relates to evolution and how method/concept/idea "dead ends" are erroneously overlooked. One thing I feel needs more attention is the fact that the external environment can change over time (moving the goal posts, if you will). What would be an evolutionary short fall in the current setup can be a defining quality in another situation allowing the model to excel. A book called "Walls: A History of Civilization in Blood and Brick" by David Frye comes to mind, where the social evolution of two fundamentally different social systems (basically sedentary and nomadic lifestyles: those inside the walls and those outside). At different stages of human development, each lifestyle could be argued was more effective than the other (the better evolutionary choice), but given a significant shift in the environment (be it natural, technological, political, economical, etc) and suddenly the advantages/disadvantages of that lifestyle changed drastically as well. I can't help but think how in our often academic ways of thinking, especially in the field of R&D, we tend to simplify our external factors in order to make sense of our progress, but in doing so we close the door on potential developments. A classic case of "opportunity costs" I guess.