Same sentence, same attention grid — rows look, columns get looked at. In Book 01 the future was hidden. Here, the mask lifts.
Click a row — a word — to see exactly which words it’s allowed to look at.
Each cell is one entry of the QKT score grid. In causal mode the upper triangle (the future) is set to −∞, so softmax crushes it to zero — the model only reads left-to-right. Drop the mask and every word reads both directions at once. That’s the whole trick behind a model that understands.