A lot of the work I have done over the past few months has somehow led me to game engines. Specifically, rendering engines using AI architectures.
In this video, he makes an interesting observation:
“The cool thing about this is that future games might not even need 3D models, they might just have latent representations of objects and use neural networks to synthesize views of them real-time. This means infinite high-fidelity graphics, with constant rendering time, that takes up very little storage space.”
I agree, since that is the general pattern of technology. More graphics, smaller memory footprint, through more efficient algorithms, aka these models.
My brother showed me the new Witcher 4 game trailer. I haven’t played many games lately, so I had no idea how far the engines had come in basically 12 months.
They are using “machine-learned deformations” (not formally programmed, aka by a programmer) to generate these images. But, they are simulating the muscles underneath, which is actually contributing to its shape. So the outer layers are inheriting math from the lower layers.
And I thought about how this could be applied to simulating a cell- can we take a similar engine from the horse, systems biology renderer, drop it down a few sizes, and use it on a cell?
How far off would it actually be?
[calculations to come at some point?]