Hacker News Reader

Not obvious to me. I thought these AI things required vast amount of GPU power. Or is that for the training part?

Inference works fine on CPUs while a bit slow but fine for a chatbot