What is an embedding, and how is it different from generated text?

Defines embeddings as model-produced vectors used for similarity and contrasts representation with text generation.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

An embedding is a vector of numbers produced by a model to represent features of an input, often so semantically related items are near each other under a chosen similarity measure. Applications use embeddings for retrieval, clustering, recommendations, or deduplication. A generative model produces a sequence such as text or tool arguments. An embedding model instead produces a representation that application code compares or stores; the vector is not a human-readable summary.

Full answer below

Practice focus

A clear conceptual boundary between embedding models, generative models, similarity, and the application decisions around retrieval.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
Embeddings represent content for comparison; they are not hidden prose, a database answer, or a generated response.
An embedding is a vector of numbers produced by a model to represent features of an input, often so semantically related items are near each other under a chosen similarity measure. Applications use embeddings for retrieval, clustering, recommendations, or deduplication. A generative model produces a sequence such as text or tool arguments. An embedding model instead produces a representation that application code compares or stores; the vector is not a human-readable summary.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
For document search, I can embed document chunks and a user's query, retrieve nearby vectors, then give the selected text—not the vectors—to a generative model as evidence. Similarity does not prove factual agreement or authorization. Results depend on the embedding model, input formatting, chunking, metadata filters, distance function, and domain. I evaluate retrieval on representative queries rather than assuming semantic search automatically understands the product.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Calling an embedding a compressed paragraph that can be read back directly.
Treating high similarity as proof that two statements are true.
Changing embedding models without an index migration plan.
Ignoring metadata filters and source authorization.

What is an embedding, and how is it different from generated text?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue practice

Next interviewer turn

Tech stacks

Skill areas

Human editorial direction

Was this drill useful?

Found this answer useful?