I built a neural network using DOOM games
Summary
The article highlights the Doomformer project, which runs neural networks inside DOOM game environments to perform exact, bit-level computations. It showcases MNIST experiments, transformer-like models, and multiple large-scale demonstrations, all implemented in an open-source framework with Python and PyTorch tooling. The piece emphasizes the novelty of integrating AI workloads into a game engine and provides links to research artifacts, demos, and tutorials.