I do want to try this line of stuff https://github.com/CASIA-IVA-Lab/FastSAM / https://github.com/YavorGIvanov/sam.cpp but I don't want to be dependent on PyTorch and Python and CUDA and 1000 other things + I don't want to be dependent on an M-series Mac CPU either, I want to like run it on the GPU on a Pi 4 or 5 or on a $150 mini-PC

i'm making a new version of the live USB that includes Nvidia drivers

is the intuition that you should use custom shaders / SDF rendering when your dataset is denser than the screen (maps, plots with millions of points, big graphs, ZUIs) -- keep all the data on GPU, only touch what pixels on the screen actually need -- and use traditional canvas-style 2D rendering when your dataset is sparser than the screen?

I guess the dream is that you could have single-use memory addresses for immutable data, where once an object is allocated at an address, you have a guarantee that the address will always refer to that object https://social.omar.website/@omar/statuses/01HX5DRDAZ3YFN7NEEX7NK93YY

Been making a "Programming" are.na https://www.are.na/omar-rizwan/programming-rvjntbotxvk

rebooting your computer because you ran out of address space

Physical computing idea (this is how I keep thinking I'd use NFC, have a set of many readers / tap points that have specific meanings)

so much of using any tool on the computer is just about struggling to select the object that you want to operate on

I was worried there was a thermal issue but it turned out it was just running out of memory :-)