What Are Vector Embeddings? (And Why Your Spotify Wrapped Knows You Too Well)

Imagine a postal worker who never learned to read. Not a single word. Can't tell an A from a Z, wouldn't recognize their own name on a birthday card. And yet, this worker has memorized the precise physical location of every house in an infinite city. They navigate by pure spatial memory, knowing exactly which homes sit in the same cul-de-sac, which ones are clear across town, and which are practically next-door neighbors. They've never read a street name or house number in their life, but ask them which residences are similar and they'll tell you instantly based on coordinates alone.

This is how vector embeddings work. An embedding is a representation of data (a word, a song, an image, anything) as a list of numbers that captures its relationships to other data. Your Spotify playlist, that photo of your dog, the word "pizza," they all get converted into coordinates in a vast mathematical space. The system doesn't "understand" content the way you do. It just knows where everything sits and can measure distances between points. Close together means similar, far apart means different.

How the Worker Learned the Territory

The worker didn't start with this comprehensive mental map. They built it gradually by walking millions of routes and noticing what appeared together. Which houses had mail delivered on Tuesdays. Which residents waved to each other. Which blocks had similar holiday decorations. Over time, patterns emerged, and the worker positioned each house based on these observed relationships.

The AI does the same. It processes massive amounts of examples and notices what appears in similar contexts. Words that show up near the same other words get placed close together in the coordinate system. "King" and "queen" both appear frequently alongside "royalty," "throne," "crown," and "castle" in text, so their coordinates land in the same neighborhood. "Dog" and "puppy" show up in similar sentences ("walked the...," "fed the...," "adopted a..."), so despite sharing zero letters, their positions sit remarkably close.

When Spotify analyzes millions of songs, each track becomes a point in mathematical space based on tempo, instrumentation, mood, lyrical themes, and which playlists group them together. Songs that appear in similar contexts (the same workout mixes, the same rainy-day collections) land near each other. When you hit play on a track you love, Spotify finds its nearest neighbors in that mathematical space and suggests them next.

The resulting space doesn't have just two dimensions like latitude and longitude. It typically has hundreds or even thousands, each capturing different subtle patterns. One dimension might encode "is this about royalty," another "is this typically young or old," another "does this appear in formal or casual writing." The worker's mental model is unfathomably complex, tracking relationships across more directions than human brains can visualize.

Google's search trained on billions of web pages this way, gradually learning which words and phrases appear in similar contexts across the entire internet. The system never attended English class, never learned grammar rules. It just walked enough routes to know the territory.

Measuring Distance Between Houses

Ask the worker which two houses are practically neighbors, and they'll tell you instantly by measuring the distance between coordinates. No need to check street names or calculate driving routes. The mathematical distance in their mental map reveals similarity directly.

Measuring the distance between two sets of coordinates (often using cosine similarity, a particular way of calculating how close two points are) tells you how similar the underlying things are. Romance novels and tax law textbooks end up in completely different neighborhoods. "Hot" and "cold" sit far apart. "Hot" and "warm" are much closer. "Boiling" and "scalding" are practically roommates.

This distance measurement enables the famous "king minus man plus woman equals queen" example. The worker can do spatial math: start at the king's house, measure the direction and distance to the man's house, then move that exact same direction and distance from the woman's house. You'll land very near the queen's coordinates. The mathematical space captured a pattern (gender relationships in royal terminology) without anyone explicitly teaching it that rule.

Netflix uses this constantly. Your viewing history gets converted into coordinates. The system measures distances to millions of other users' coordinates, finds your nearest neighbors in taste-space, then recommends what those nearby viewers watched. Nobody manually tagged "Breaking Bad" as similar to "Better Call Saul." The coordinates just landed close together because people who watched one often watched the other.

Why This Illiterate Postal Worker Is Everywhere Now

The same trick works for any kind of data. Turn it into coordinates, measure distances, find neighbors. Product descriptions become coordinates powering Amazon's recommendations. Customer support tickets become coordinates that automatically route to the right team. Images become coordinates, which is how Instagram recognizes faces and how Google Photos finds every picture of your cat without you tagging a single one.

The worker doesn't need to understand what makes two houses similar, just that their coordinates sit close together. The AI doesn't need to "know" why two songs feel related or why two faces look alike. The mathematical distances handle everything.

This spatial approach also enables comparing across different types of content. Text search can find relevant images because both get converted to coordinates in related spaces. You can describe a mood in words and get song recommendations because the coordinates capture abstract similarities like "energetic" or "melancholy" that transcend the specific medium.

Your iPhone's photo app turns every face into coordinates, then clusters nearby points together. Those 47 photos spanning five years, different lighting, different angles, different haircuts? All land close enough in face-space that the system realizes they're the same person and creates an album automatically. You never labeled anything. The distances told the whole story.

So What Can YOU Do With This?

Understanding coordinates helps you use AI search tools more effectively. Instead of hunting for exact keywords, search by concept or example. Many new tools let you search your own documents, photos, or bookmarks by describing what you want ("that article about productivity I read last spring") rather than remembering precise titles.

You can improve your prompts to ChatGPT by knowing it finds responses "near" your query in meaning-space. More context helps it triangulate your location in that space and find better neighbors. Vague queries leave it guessing which neighborhood you're even in.

If you're building anything involving recommendations or search, coordinate-based tools are now accessible without advanced degrees. Vector databases like Pinecone, or built-in features in Notion and Obsidian, let you add "find similar items" functionality to your projects. You feed in your data, the system converts it to coordinates, and similarity search becomes almost trivial.

This knowledge also clarifies why AI makes certain weird mistakes. If two unrelated things accidentally land near each other in the mathematical space (maybe they appeared in similar contexts by coincidence), the system thinks they're similar even when they're obviously not to a human. The worker trusts the coordinates completely, even when the map occasionally lies.

Google Photos lets you search "beach sunset" and surfaces those exact scenes despite zero manual tags. ChatGPT answers questions phrased awkwardly because it finds meaning-neighbors to your garbled query. Both rely on placing related concepts near each other in coordinate space, making "close enough" astonishingly powerful.

TL;DR

Vector embeddings convert any data (words, images, songs) into coordinate lists where mathematical distance represents similarity, powering most modern AI recommendations and search. AI learns these coordinates by processing millions of examples, placing items that appear in similar contexts near each other automatically without manual rules. This single technique drives Netflix recommendations, Google Photos face recognition, Spotify suggestions, and semantic search across basically every platform you use daily. You can use coordinate-powered search right now in tools like Google Photos, Notion, and ChatGPT to find things by describing what you want instead of remembering exact keywords.

The postal worker still can't read their own name tag, but they just delivered your package to the right house on the first try anyway.