SnowPro Specialty: Gen AI · 24% of the exam

Snowflake for Gen AI Overview: free practice questions

5 sample questions from our 75-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Snowflake exam.

1. A stakeholder asks whether Snowflake trains its base models on the customer's prompts and data submitted to Cortex. What is the correct understanding to convey?

  • A. Customer data submitted to Cortex is not used to train the underlying foundation models✓ Correct
  • B. All customer prompts are automatically added to the next model's training set
  • C. Cortex publishes customer data to a public dataset by default
  • D. Customers must opt out in writing every day to avoid training
Explanation

Snowflake's Cortex service does not use customer data to train the foundation models; data stays within the customer's governance boundary. The other options misstate Snowflake's data-handling posture.

2. A finance analyst wants to type 'what was Q2 revenue by region' in plain English and get a computed numeric answer sourced from the sales tables, without writing SQL. Which Cortex capability should be used?

  • A. Cortex Analyst✓ Correct
  • B. Cortex Search
  • C. AI_COMPLETE called directly on the raw table rows
  • D. Snowflake CoCo
Explanation

Cortex Analyst converts natural language into SQL against a governed semantic model to compute answers from structured tables, which is exactly this scenario. Cortex Search retrieves text, not computed aggregates. Calling AI_COMPLETE directly on rows would not reliably compute an aggregate like revenue by region. CoCo is a coding/admin assistant, not a business-question tool.

3. A hospital wants to let clinicians summarize patient notes using generative AI, but compliance requires that no patient data ever be transmitted to an external LLM API. What property of Snowflake Cortex makes it a fit here?

  • A. Cortex inference runs inside the account's governance boundary, so prompts and data are not shipped to a third-party LLM API✓ Correct
  • B. Cortex automatically encrypts all data before sending it to the model provider's API
  • C. Cortex only supports models that have signed a data-processing addendum with the hospital
  • D. Cortex requires the hospital to host the model themselves on-premises
Explanation

Cortex's defining governance property is that inference happens inside the Snowflake account boundary rather than being shipped to an external LLM API. Encryption-in-transit to a third party is not the relevant guarantee here; Cortex doesn't require a DPA with a model provider because there is no external transmission; and it does not require self-hosting on-premises, which describes the opposite of managed Cortex.

4. When selecting a model for a Cortex COMPLETE call, which factors should most directly drive the choice?

  • A. The task's required quality, latency tolerance, context-window needs, and cost per token✓ Correct
  • B. The color scheme of the Snowsight worksheet
  • C. The number of columns in the source table
  • D. The alphabetical order of the model names
Explanation

Model choice is a trade-off among answer quality, latency, context window, and cost. Larger models generally cost more and may be slower; smaller models are cheaper and faster for simpler tasks. The other options are irrelevant.

5. A scenario states a team wants to 'let a business user get a plain-English answer that involves both looking something up in a document and computing a number from a table, without the user knowing which tool did which part.' What does the phrase 'without the user knowing which tool did which part' signal?

  • A. The need for orchestration across tools, pointing to Cortex Agents rather than exposing Search and Analyst as two separate, user-facing steps✓ Correct
  • B. The need for a resource monitor to hide compute costs from the user
  • C. The need for row access policies to hide table rows from the user
  • D. The need for Cortex Guard to filter which tool names are shown
Explanation

Hiding tool selection behind a single conversational interface that spans multiple capabilities is exactly the orchestration role Cortex Agents play. A resource monitor governs credit spend, not user-facing tool transparency. Row access policies restrict which data rows a role can see, unrelated to hiding orchestration mechanics. Cortex Guard filters unsafe content in responses; it has no role in concealing which tool was used.

70 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free
Snowflake for Gen AI Overview — Free SnowPro Specialty: Gen AI Practice Questions | DataCertPrep — Certification Prep