Before a language model can reason about text, it must convert words (or sub-word tokens) into numbers. Each word is mapped to an embedding vector whose direction and position encode meaning — words with similar meanings end up near each other in that vector space. This 2D companion keeps the same hand-placed coordinates and cosine-similarity math as the 3D version, just projected onto a flat plane you can pan and zoom directly.
- Clusters — words from the same semantic category (animals, emotions, tech, royalty, food) sit close together, exactly as real embeddings from models like Word2Vec or GPT cluster synonyms and related concepts.
- Neighbor links — a line is drawn between two words when the cosine similarity of their vectors passes your chosen threshold:
cos(a,b) = (a·b) / (|a||b|).
- Analogy vector — toggling this shows the classic
king − man + woman ≈ queen relationship: the same directional "royalty" offset applied to a different starting point lands near a semantically related word.
- Click any word to highlight its closest neighbors; drag to pan, scroll/pinch to zoom.