https://webcyte.io has a custom implementation of some of these algorithms - allows you to study the code and hack on it, while running everything in the browser… it’s my personal project, work-in-progress, but may be useful for people who learn from code, like myself.
I was doing something similar with fluid equations on meshes more than ten years ago. Once you have Automatic Differentiation it's fairly straightforward to reformulate a time evolution process into one that solves for some free parameter.
It's a pity they don't give a video of the butterfly battle, that sounds a lot more impressive than the static 2d lizard.
What I'd love to see is a reproduction of the kind of embryo development videos you can get from lightsheet microscopy, e.g. https://youtube.com/watch?v=2Vnyph3Vmic
Thanks. Starting with the butterfly and playing with the rotation slider I did manage to produce some seriously "biblically accurate" butterflies. No emergent evolutionary biology yet, but still fun.
You may also like this how this mostly hand-designed CA rule can produce plausible mutations when disturbed: https://www.youtube.com/watch?v=qwJNeq-WABU ("From One Cell to a Multicellular Organism", Part 2, 25min, Simulife Hub)
Well this will support more of that simulation theory supporters, similar to James Gates claims that you can find computer code in the equations that we use to describe our universe in string theory.
Yes we do control much more complex systems and optimize much more complex objectives than just a target arrangement of pixels. It is an optimization problem that people were trying to solve with heuristics. Leetcode style.
Of course it was hard if not impossible. And of course even a suboptimal neural network can do it easily.
Readit News
It's a pity they don't give a video of the butterfly battle, that sounds a lot more impressive than the static 2d lizard.
What I'd love to see is a reproduction of the kind of embryo development videos you can get from lightsheet microscopy, e.g. https://youtube.com/watch?v=2Vnyph3Vmic
Of course it was hard if not impossible. And of course even a suboptimal neural network can do it easily.