Building my own Diffusion Language Model from scratch was easier than I thought

Since I felt like I was relying on Claude Code a lot recently, I wanted to see how hard it is to implement a diffusion language model from scratch without the help of AI-Generated code. So I built one while waiting for the training for my master's thesis. This is what I got after a few hours of training on my MacBook Air M2. I trained on the tiny Shakespeare dataset from Karpathy and prompted "to be, " To be, fo hend! First her sense ountier to Jupits, be horse. Words of wisdom! The model ha