Embedding Space
Type words and watch them scatter. The first word anchors the center, the rest land by cosine similarity.
A language model holds every word it knows in a high-dimensional space. Nearby points mean nearby meanings. We never get to stand inside that space - it has hundreds of dimensions and no doors. These three small games are doors.
Each one takes the same hidden machinery - sentence embeddings running entirely in your browser - and turns it into something you can touch. One maps words by how close they sit. One hides a word and lets you hunt it by warmth. One asks a bot to remember what you feed it, and watches it fumble. Together they are a way of playing with how a machine thinks it understands us.
Type words and watch them scatter. The first word anchors the center, the rest land by cosine similarity.
Hunt a hidden word by meaning. Semantic distance is the sonar, and warmer guesses sink the ship.
Feed a bot memories and talk to it. A small study of machine memory and how clumsily it recalls.