Knowledge Tracing Machines (KTMs) recover Multidimensional Item Response Theory (MIRT) when the embedding dimension is strictly positive ($d > 0$). Whereas $d = 0$ (no embeddings) reduces a KTM over users and items to unidimensional IRT, a strictly positive $d$ assigns each user and item a $d$-dimensional embedding whose interaction represents multiple latent abilities and difficulties—the multidimensional generalization of IRT that MIRT provides.