Why can a machine recognize that "cancel my reservation" and "call off my booking" are close in meaning even when the words are different? The answer begins with one deceptively simple idea: similarity.Similarity: How AI Finds Things That Mean Something Similar is a first-principles guide for readers who want to understand semantic similarity without being buried in jargon, programming, or unexplained mathematics. It begins with comparisons you already make in everyday life, then builds carefully toward vectors, embeddings, distance, dot product, normalization, cosine similarity, ranking, thresholds, top-k retrieval, reranking, and evidence-aware decisions.The goal is not to make you memorize formulas. It is to help you understand what each formula is trying to measure, what a similarity score can actually tell you, and where that score can mislead you.>- separate "same," "similar," "related," "relevant," and "useful"- understand vectors and embeddings as representations rather than mysteries- read cosine similarity from first principles without fear of mathematics- distinguish ranking from confidence, thresholds, and final decisions- recognize hard negatives, contradiction, stale evidence, ambiguity, and near-ties- understand why freshness, authority, scope, permissions, and exact details may matter alongside semantic closeness- break complex retrieval and comparison problems into smaller, solvable questions This is a book about more than one metric. It is about learning how to think when an AI system says two things are "close." By the end, you will be able to question a similarity system intelligently, trace how a match becomes a shortlist, and see why the nearest result is only a candidate-not a verdict.If embeddings, semantic search, retrieval, vector mathematics, or cosine similarity have ever felt like hidden machinery, this book gives you a floor to stand on before the next layer begins.
