> Start here if you're new to ML. In nineteen problems we'll get from dot products to a working training loop, all in NumPy. Implement the building blocks (sigmoid, softmax, cross-entropy, attention masks), then build a linear layer and work through the backwards passes that make training work.