The AI community has already seen multiple attempts at visualizing model internal structures, such as heatmaps of Transformer attention mechanisms and dynamic demonstrations of neural network activations.