AIHacker Newsabout 2 hours ago
Developer Compiles DOOM into Transformer
18 min read
A developer released Doomformer, a project compiling integer neural networks into ZDoom maps. The system runs on ViZDoom and uses 1,355 voodoo dolls to compute 784 pixels. It achieves 99 correct reads on 100 MNIST test digits and 966 correct reads on 1,000 sampled digits. The tool requires Python 3.11 and Linux or macOS, with installation via pip or git.
Level
Hype check
Tap to vote and see what everyone thinks.