An embedding maps an input such as text or an image into a numerical representation. A search system can compare these representations to identify items that may be relevant to a query, even when their wording differs.
Similarity depends on the model, the data and the comparison method. An embedding can help retrieve an item, but it does not by itself interpret every detail of the source or generate a reliable answer about it.
ELI5
An embedding gives information a numerical description that a computer can compare with other descriptions. Items with related meaning can have descriptions that are close together.
For example, a search for 'leaking tap' may find a maintenance note that says 'dripping faucet'. The matching descriptions help find the note, but the assistant still needs to read it before explaining the repair history.
Does an embedding generate an answer?
Not by itself. It represents information for tasks such as search, while answer generation is a separate step.
Does similarity prove that two items mean the same thing?
No. It is a model-dependent comparison that can return useful but imperfect matches.
