the new 4x4 matrix multiplication is interesting though... there are a lot of usecases for this

General improvements, some not

Full video with a bunch of filler

>>3860
this video is actually 2 years old and this newer method applies more generally

https://archive.is/NBaXE
Also like optimizing their cloud cluster to come up with better organization to save 0.7% on their costs... It's not being released to the public but it's a pretty impressive statement.

I have no idea what I'm looking at except from some hexagons?

>>3864
Here, read this
https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf

>>3865
I don't think I wanna...

I can't run the notebooks, but this has some information I think on how the 48 step 4,4,4 matrix multiplication works
colab.research.google.com/github/google-deepmind/alphaevolve_results/blob/master/mathematical_results.ipynb#scrollTo=nNmZSa-oeJ2O

>>3866
Games use 4x4 matrices to represent things, and when the camera moves it has to apply 4x4 matrices against everything.

I don't really understand how this system works or if their are limitations. But it seems to be implying that that GPUs can run 3D simulations faster because of a math find that one of Google's AIs did, using a training system that allows them to feed back the results of code into itself to make improvements.

Pretty much just statements though.