This is true but only at present. An analogy would be the weights of pretrained machine learning models. They're still serializable, though it's true they're l

This is true but only at present. An analogy would be the weights of pretrained machine learning models. They're still serializable, though it's true they're less semantically trackable than a well-written TikZ snippet (which, by the way, I haven't seen many of in practice!)