Course contents

SnowPro Core COF-C03 Study GuidePart 7: AI featuresPractice Questions: Part 7

Practice Questions: Part 7

Practice Questions · Part 7

Thirty questions in the exam's format: single choice and multiple select, no topic hints. Answer each one, then read the explanation even when you got it right.

Practice questions

  1. A support team wants every ticket labeled with one of five product areas, which it supplies as a list. The team doesn't want to write a prompt or choose a model. Which function fits?

    1. AI_CLASSIFY with the five areas passed as categories
    2. AI_EMBED, then AI_SIMILARITY against each area
    3. AI_SENTIMENT with the five areas passed as categories
    4. AI_COMPLETE with a prompt that lists the five areas
  2. In an account that still has its default settings, any role can call AI_COMPLETE. Which two grants held by PUBLIC make this possible? (Choose two.)

    1. The USE AI FUNCTIONS account-level privilege
    2. The SNOWFLAKE.CORTEX_USER database role
    3. IMPORTED PRIVILEGES on the SNOWFLAKE database
    4. The SNOWFLAKE.AI_FUNCTIONS_USER database role
    5. The SNOWFLAKE.CORTEX_EMBED_USER database role
  3. A team replaces SNOWFLAKE.CORTEX.SENTIMENT with AI_SENTIMENT in a dashboard query. The dashboard expected a number between -1 and 1. What does AI_SENTIMENT return instead?

    1. An integer rating from 1 to 5 for each input text
    2. The same FLOAT score, so the dashboard needs no change
    3. An object whose categories carry labels such as positive
    4. A VECTOR value to compare with AI_SIMILARITY
  4. A query running on warehouse BI_WH calls AI_COMPLETE once for each of 10,000 rows. Which charges does this produce?

    1. Warehouse credits only, since the model runs on the warehouse
    2. Warehouse credits while it runs, plus AI credits for the tokens
    3. AI credits only, since the warehouse pauses during each call
    4. A flat fee for each call, whatever the length of the text
  5. A nightly job runs AI_CLASSIFY over 2 million rows on an X-Large warehouse. What does Snowflake recommend for the warehouse?

    1. Convert it to a Snowpark-optimized warehouse for memory
    2. Make it multi-cluster so that rows spread across clusters
    3. Use a warehouse no larger than Medium for AI functions
    4. Resize it to 4X-Large so the model calls finish sooner
  6. Which two statements about how Cortex AI Functions are billed are correct? (Choose two.)

    1. AI_CLASSIFY bills only input tokens, since it returns a label.
    2. AI_COMPLETE counts both input and output tokens.
    3. A resource monitor can suspend AI spend when its quota is reached.
    4. AI_EMBED counts only input tokens.
    5. Token charges are added to the warehouse's per-hour credit rate.
  7. A FinOps lead wants an email when the account's total spend for the month, including AI functions and other serverless features, is projected to exceed 20,000 credits. Which feature fits?

    1. A daily query on WAREHOUSE_METERING_HISTORY with an alert
    2. An account-level resource monitor with a 20,000-credit quota
    3. The account budget, activated with a 20,000-credit limit
    4. A resource monitor on every warehouse with a NOTIFY trigger
  8. A regulated company requires that its prompts to AI functions be processed only in the account's home region. What should the account administrator set?

    1. CORTEX_ENABLED_CROSS_REGION = 'DISABLED'
    2. CORTEX_ENABLED_CROSS_REGION = 'AWS_US'
    3. CORTEX_ENABLED_CROSS_REGION = 'ANY_REGION'
    4. A network policy that blocks the other regions
  9. Where do the large language models behind Snowflake's AI functions run when a call to AI_COMPLETE is processed?

    1. In a cloud account that the customer provisions for inference
    2. Inside the Snowflake service perimeter, other vendors' models included
    3. On the model vendor's public API, reached over the internet
    4. On the nodes of the warehouse that runs the calling query
  10. A company stores 50,000 help-center articles as text in a table. It wants a chat app that finds the most relevant passages for each user's question before an LLM writes the answer. Which feature provides the retrieval step?

    1. Cortex Analyst with a semantic view over the article table
    2. The search optimization service with SUBSTRING on the text
    3. A Cortex Search service on the article text column
    4. AI_AGG over the article column, called for each question
  11. Which three are cost components of a Cortex Search service? (Choose three.)

    1. Nothing at all in months when no one queries the service
    2. Embedding tokens for rows that are added or changed
    3. A charge for each successful message the service answers
    4. Serving compute per GB per month of indexed data
    5. Warehouse credits to build and refresh the index
  12. Role SUPPORT holds USAGE on Cortex Search service KB_SEARCH and on its database and schema, but no privilege on the source table. A row access policy on that table would hide half of its rows from SUPPORT. What can SUPPORT's searches return?

    1. An error, because SUPPORT can't read the source table
    2. Nothing, until SUPPORT gets SELECT on the source table
    3. Matches from any row the service has indexed
    4. Only matches from rows the policy lets SUPPORT see
  13. What does TARGET_LAG = '1 hour' control on a Cortex Search service?

    1. How long search results are cached for repeated questions
    2. How often the owner is billed for the serving compute
    3. How far the indexed content may trail the base data
    4. The longest time a search may run before it's cancelled
  14. How is Cortex Analyst billed?

    1. Per GB per month of the tables the semantic view covers
    2. Per successful message, plus a warehouse to run its SQL
    3. Per token in the semantic view, read on every question
    4. Only through the warehouse; Analyst itself costs nothing
  15. A team is starting a new Cortex Analyst project. Where does Snowflake recommend declaring the tables, joins, metrics and synonyms that Analyst uses?

    1. In a semantic view created in a database schema
    2. In a YAML semantic model file uploaded to a stage
    3. In the ATTRIBUTES clause of a Cortex Search service
    4. In comments on each table and column it should read
  16. Given privileges on the underlying objects, which two database roles each let a role use Cortex Analyst? (Choose two.)

    1. SNOWFLAKE.CORTEX_USER
    2. SNOWFLAKE.GOVERNANCE_VIEWER
    3. SNOWFLAKE.CORTEX_ANALYST_USER
    4. SNOWFLAKE.COPILOT_USER
    5. SNOWFLAKE.USAGE_VIEWER
  17. Analysts with role EU_FIN use a Cortex Analyst app that runs the generated SQL in the analyst's own session. Table REFUNDS has a row access policy that returns only European rows to EU_FIN. An analyst asks for total refunds by region. What do the results include?

    1. Every region, since the semantic view grants the access it needs
    2. European regions only, since the caller's role and policies apply
    3. An error, since policies block SQL that Analyst has generated
    4. Every region, since Analyst runs its SQL with the owner's rights
  18. What does Cortex Analyst use to generate SQL for a user's question?

    1. The account's query history, mined for similar past questions
    2. A sample of rows that it reads from each table in the model
    3. The full table contents, embedded into a hybrid search index
    4. The semantic model's metadata: names, types and descriptions
  19. A pipeline calls invoice_model!PREDICT on a Document AI model build to pull vendor, amount and due date from PDF invoices. What is Snowflake's documented replacement?

    1. AI_EXTRACT, with the fields in its responseFormat
    2. AI_PARSE_DOCUMENT in LAYOUT mode on each file
    3. AI_CLASSIFY, with the fields passed as categories
    4. A Cortex Search service over the invoice stage
  20. How does a data engineer tell AI_EXTRACT which values to pull from each document?

    1. By labeling sample documents in a model builder, then publishing it
    2. By passing a regular expression for each field in the function call
    3. In the responseFormat argument, as named questions or a JSON schema
    4. By creating a semantic view that maps each field to a target column
  21. Which two pairings of a job and a Snowflake feature are correct? (Choose two.)

    1. Let finance users ask revenue questions over tables: Cortex Search
    2. Return the best passages from a policy library: Cortex Search
    3. Project next quarter's daily order counts: ANOMALY_DETECTION
    4. Pull vendor and total from PDF invoices in a stage: AI_EXTRACT
    5. Find a table a colleague mentioned by name: Cortex Analyst
  22. A team wants the full text of scanned contracts, with headings and tables kept in reading order, so it can split them into chunks for a Cortex Search service. Which function fits?

    1. AI_PARSE_DOCUMENT in LAYOUT mode on each contract
    2. AI_PARSE_DOCUMENT in OCR mode on each contract
    3. AI_TRANSCRIBE on each scanned page of the contracts
    4. AI_EXTRACT with one question for each heading
  23. Which two statements about Cortex Code (CoCo) in Snowsight are correct? (Choose two.)

    1. It can be used without any Cortex database role.
    2. It's the REST API that answers questions over semantic views.
    3. It performs only the actions the active role permits.
    4. It takes over from Snowflake Copilot, which is deprecated.
    5. It only drafts SQL; it can't run statements or change objects.
  24. Which statement about Snowflake's ML functions, such as SNOWFLAKE.ML.FORECAST, is correct?

    1. They're serverless and billed per million tokens, like AI_COMPLETE
    2. They need a Cortex Search service to store each trained model
    3. Training and prediction run on a virtual warehouse that you provide
    4. They can be called only through the Cortex Analyst REST API
  25. An operations team wants each day's order count flagged when it falls outside the range its history predicts, with a lower and an upper bound for each day. Which feature fits?

    1. SNOWFLAKE.ML.CLASSIFICATION
    2. AI_FILTER over the order table
    3. SNOWFLAKE.ML.ANOMALY_DETECTION
    4. SNOWFLAKE.ML.FORECAST
  26. Training an ANOMALY_DETECTION model on a single series of 40 million rows runs out of memory on a standard warehouse. What does Snowflake suggest?

    1. Label anomalies with AI_COMPLETE instead
    2. Make the standard warehouse multi-cluster
    3. Train it on a Snowpark-optimized warehouse
    4. Enable query acceleration on the warehouse
  27. Which two statements about AI_AGG are correct? (Choose two.)

    1. It requires a model name as its first argument.
    2. It fails when a group's text exceeds one model context window.
    3. It takes an instruction you write, such as "List the top complaints".
    4. It returns one string for each group, so it's used with GROUP BY.
    5. It returns TRUE or FALSE for each row it reads.
  28. Which statement about AI_FILTER is correct?

    1. It returns one string per group of rows, like an aggregate
    2. It requires a model name as its first argument, like AI_COMPLETE
    3. It returns a BOOLEAN, so it fits in a WHERE clause or a join condition
    4. It returns an object with a labels array, like AI_CLASSIFY
  29. Which two functions require the caller to name a model? (Choose two.)

    1. AI_SENTIMENT
    2. AI_FILTER
    3. AI_COMPLETE
    4. AI_TRANSLATE
    5. AI_EMBED
  30. A FinOps analyst needs last month's AI function credits broken down by function, model and user. Which source answers this?

    1. SNOWFLAKE.ACCOUNT_USAGE.QUERY_ACCELERATION_HISTORY
    2. SNOWFLAKE.ACCOUNT_USAGE.CORTEX_AI_FUNCTIONS_USAGE_HISTORY
    3. SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
    4. SNOWFLAKE.ACCOUNT_USAGE.METERING_DAILY_HISTORY