Why AI Search Beats Keyword Search (Embeddings Explained)

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-foundations-ai-embeddings-vector-search-r9a10-teen

1. What is an embedding in the context of AI?

   1. A search algorithm that only finds exact keyword matches
   2. A tool that translates text between different languages
   3. A list of about 1,500 numbers that represents the meaning of a piece of text
   4. A type of database that stores complete copies of documents

2. Which technology do most 'chat with your documents' applications use under the hood?

   1. Regular expressions and keyword matching
   2. A manual review process where humans read every query
   3. A simple alphabetical sorting system
   4. Embeddings combined with vector search

3. What is a vector database designed to do?

   1. Store images and display them on websites
   2. Store embeddings and quickly find other embeddings that are most similar
   3. Store complete audio files and play them back
   4. Store only text files in alphabetical order

4. Spotify's recommendation system works by giving each song an embedding. What happens to songs with nearby embeddings?

   1. They are always played in alphabetical order by title
   2. They get recommended to the same listeners because they are similar
   3. They are translated into different languages
   4. They are deleted from the platform automatically

5. In Notion AI search, a query for 'find my notes about anxiety' matches notes containing 'stress' or 'overwhelmed'. Why does this work?

   1. Because embeddings capture meaning, so 'anxiety,' 'stress,' and 'overwhelmed' have similar vector representations
   2. Because the notes were manually tagged by humans with every possible emotion word
   3. Because the search reads the user's mind to know what they really mean
   4. Because Notion has a built-in dictionary that manually maps every emotion to every synonym

6. What does it mean mathematically when two embeddings are 'close together'?

   1. The documents were written at approximately the same time
   2. The numbers in their lists are similar, suggesting the texts have similar meanings
   3. The texts use many of the same individual words
   4. The files are stored on the same computer

7. What problem do embeddings solve that keyword search cannot solve?

   1. Understanding that different words can have the same meaning (synonyms)
   2. Finding documents that contain every word in the query
   3. Sorting results alphabetically
   4. Counting how many times each word appears in a document

8. What does RAG stand for in AI?

   1. Random-Alpha-Gradient
   2. Recursive-Array-Graph
   3. Read-All-Generate
   4. Retrieval-Augmented Generation

9. If you wanted to build a feature that lets users ask questions about your private PDF documents, what combination of technologies would you most likely use?

   1. A keyword-only search engine like those from the 1990s
   2. A simple spreadsheet to list all document titles
   3. A manual lookup system where humans answer every question
   4. Embeddings to represent document chunks and a vector database to find relevant ones

10. What is the main advantage of using embeddings for search compared to traditional keyword matching?

    1. Embeddings can search through images but keywords cannot
    2. Embeddings always return results faster than keyword search
    3. Embeddings can find results that use different words but have similar meaning
    4. Embeddings require less computer memory than keyword indexes

11. Why is the term 'semantic search' used to describe AI-powered search?

    1. Because it uses semicolons in its code
    2. Because it only searches through semantic websites
    3. Because it focuses on the meaning (semantics) of words rather than exact matches
    4. Because it was invented by a company called Semantic

12. The lesson mentions you can run an embedding example on a CSV of your own notes. What would you be searching for after converting your notes to embeddings?

    1. Notes that are stored in the first row of the CSV
    2. Notes that contain the exact same words as the query
    3. Notes with the most similar meaning to a query
    4. Notes that were written most recently

13. What is the mathematical nature of an embedding?

    1. A list of hundreds or thousands of numbers representing a point in space
    2. A physical location where files are stored
    3. A color code that can be displayed on screen
    4. A single word that summarizes a document

14. Which of these is a real vector database mentioned in the lesson?

    1. Amazon
    2. Pinecone
    3. Microsoft Word
    4. Facebook

15. The lesson says embeddings let AI 'know what you mean.' What specifically allows this to happen?

    1. The AI always asks the user to clarify exactly what they meant
    2. The training process creates numerical representations where similar meanings end up numerically close
    3. The search system checks a dictionary of all possible definitions
    4. The AI reads the user's thoughts directly from their brain waves