SnowPro Associate: Platform Question 331
Single answer● COMPLETE functionA data engineering team wants to add a natural-language helper to an internal analytics app in Snowflake. They plan to call the COMPLETE function from SQL to summarize support tickets and generate short responses for analysts. The security team requires that the implementation minimize the risk of exposing unrelated database content and keep the model focused only on the text supplied in each query. Which approach best meets this requirement?
- A
Call COMPLETE with the ticket text explicitly included in the prompt and clear instructions, so the model response is grounded only in the provided input.
- B
Grant the role running COMPLETE broader SELECT access to all schemas, because the function needs to read Snowflake tables directly to improve response quality.
- C
Create a dynamic table containing all internal documents and assume COMPLETE automatically uses it as context without mentioning the content in the prompt.
- D
Use COMPLETE without providing ticket text, because the function automatically discovers relevant rows from the current database and schema.
Show answer and explanation
Correct answer: A
Explanation
The COMPLETE function is used to send a prompt to a supported model and return generated text. For secure and predictable results, applications should explicitly provide the relevant context in the prompt rather than assuming the model can access Snowflake tables, schemas, or other account data on its own. This is especially important when handling sensitive enterprise data: use least privilege for the executing role and include only the necessary ticket text or summarized context in the prompt. Snowflake documentation for Cortex AI functions emphasizes supplying the prompt input directly and managing access through standard Snowflake privileges and application design rather than relying on implicit model access to database objects.
- A. Correct.
Correct. COMPLETE generates a response from the model based on the prompt you send. If the goal is to limit exposure and keep output focused, the safest design is to pass only the specific ticket text and task instructions required for that request. This aligns with prompt-grounding best practices: include only the minimum necessary context in the prompt.
- B. Incorrect.
Incorrect. COMPLETE does not need broad table privileges to magically inspect account data. Snowflake access control still applies, and overgranting privileges increases risk rather than improving model quality. The model only works with the input provided to the function call unless your SQL explicitly retrieves and includes additional content.
- C. Incorrect.
Incorrect. A dynamic table does not become automatic context for COMPLETE. Snowflake will not implicitly attach table contents to the model prompt. If you want the model to use document content, your SQL or application logic must explicitly retrieve and provide that content.
- D. Incorrect.
Incorrect. COMPLETE does not automatically search the current database, schema, or query history for relevant information. A common misconception is that LLM functions have ambient access to Snowflake objects. In practice, the function responds to the prompt text you provide.