SnowPro Specialty: Gen AI Question 84
Single answer2.1 Apply Gen AI and LLM functions in Snowflake.A support analytics team stores customer case notes in a Snowflake table and wants to add an automatically generated one-sentence summary for each note directly in SQL. The team wants the fastest implementation using Snowflake-managed LLM capabilities and does not want to build or host a model outside Snowflake. Which approach best meets these requirements?
- A
Use the SNOWFLAKE.CORTEX.SUMMARIZE function in a SQL statement to generate summaries from the case note text column.
- B
Export the case notes to an external Python service, call an open-source summarization model, and write the results back to Snowflake.
- C
Create a Snowpark UDF that trains a custom summarization model inside the virtual warehouse and stores the model weights in an internal stage.
- D
Use a standard SQL aggregation such as LISTAGG on the case note text to create a one-sentence summary for each case.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use SNOWFLAKE.CORTEX.SUMMARIZE because the scenario calls for in-platform summarization with minimal setup and no external model hosting. Snowflake Cortex provides SQL-accessible LLM functions for common generative tasks such as summarization, making it well suited for teams that want to enrich table data directly within Snowflake. In contrast, exporting data to external services increases complexity and can conflict with governance or latency goals. Traditional SQL text functions like LISTAGG are not generative and cannot replace an LLM-based summary. This aligns with Snowflake best practices for applying Cortex functions when a managed GenAI capability already exists for the task.
- A. Correct.
Correct. SNOWFLAKE.CORTEX.SUMMARIZE is a Snowflake-managed LLM function designed to generate summaries from input text directly in SQL. It matches the requirement to keep implementation inside Snowflake and avoid building, hosting, or integrating an external model service. This is the most direct and operationally simple approach for adding generated summaries to rows in a table.
- B. Incorrect.
Incorrect. Although this could work technically, it does not meet the stated requirement for the fastest implementation using Snowflake-managed LLM capabilities. Exporting data to an external service adds operational overhead, integration complexity, network movement, and governance considerations that Snowflake Cortex functions are intended to reduce.
- C. Incorrect.
Incorrect. Snowpark UDFs can execute custom logic, but training and managing a custom summarization model is not the simplest or intended solution for this scenario. The requirement is to use Snowflake-managed LLM functionality rather than build and host a model workflow. This option introduces unnecessary complexity and does not reflect the typical use of Cortex LLM functions.
- D. Incorrect.
Incorrect. LISTAGG concatenates strings; it does not understand meaning or generate abstractive summaries. Someone might choose this if they confuse text processing with generative AI, but it cannot produce a true one-sentence summary of unstructured support notes.