Advertisement
click and drag any word to move it
Words near each other = high cosine similarity. Drag to explore.
What you're seeing
Real embeddings live in 768-1536 dimensions. 2D versions (UMAP/t-SNE projections) preserve relative distances approximately.
Cosine similarity measures angle between vectors, ignoring magnitude. Standard for semantic similarity. Dot-product or L2 also used; cosine wins for length-invariance.
★ KEY TAKEAWAY
Cosine similarity = dot product of unit vectors = cosine of angle. Foundation of semantic search.
▶ WHAT TO TRY
- Drag word vectors closer/farther.
- Watch top-5 similar pairs change.
- Normalizing makes magnitude irrelevant; only direction matters.