Anonymous
6/17/2025, 5:47:04 PM
No.105622033
>>105621113
Its been decades since he made anything that anyone considers leading edge, even the stuff he did with Aerospace and VR was pretty inconsequential in the grand scheme of things. Compare that to Doom which invented a genre and, more importantly imo, Quake which still had functions that still do the vector math for UE5 and others.
Also, what he presented is not low-level at all, nothing in his wheelhouse in software. It was robotics and mostly applied classic RL to covnets. He admitted to being too low-level to start; writing C++ writing CUDA kernels directly, then using cuBLAS, then cuDNN, then finally moving to PyTorch just so he would have a working demo to experiment with. While having Carmack write a super efficient CUDA kernel sounds like something he can do and be relevant, its not what's happening. He's simply retreading ground OpenAI and Deepmind did in 2013 - while trying to see if he could do something different, so far he has not found anything novel.
Its been decades since he made anything that anyone considers leading edge, even the stuff he did with Aerospace and VR was pretty inconsequential in the grand scheme of things. Compare that to Doom which invented a genre and, more importantly imo, Quake which still had functions that still do the vector math for UE5 and others.
Also, what he presented is not low-level at all, nothing in his wheelhouse in software. It was robotics and mostly applied classic RL to covnets. He admitted to being too low-level to start; writing C++ writing CUDA kernels directly, then using cuBLAS, then cuDNN, then finally moving to PyTorch just so he would have a working demo to experiment with. While having Carmack write a super efficient CUDA kernel sounds like something he can do and be relevant, its not what's happening. He's simply retreading ground OpenAI and Deepmind did in 2013 - while trying to see if he could do something different, so far he has not found anything novel.