SnowPro Specialty: Gen AI Question 34
Single answerCortex LLM Playground (Public Preview)A data science team is evaluating several foundation models for an internal customer-support summarization use case. They want a fast way to compare how different prompts and models behave on the same sample ticket text before building anything in SQL or Python. They also need to keep the evaluation inside Snowflake rather than calling external tools. Which Snowflake feature should they use first?
- A
Use Cortex LLM Playground to interactively test prompts against supported models in Snowsight and compare responses on representative inputs
- B
Create a Snowpark Container Services deployment to host open-source models, because LLM Playground only works after a custom model endpoint is deployed
- C
Use Dynamic Tables to automatically generate summaries for each support ticket, because LLM Playground is intended only for monitoring completed inference jobs
- D
Build a Streamlit in Snowflake app immediately, because LLM Playground cannot be used to experiment with prompts before application development
Show answer and explanation
Correct answer: A
Explanation
The best answer is Cortex LLM Playground because it is intended for experimenting with prompts and supported LLMs directly in Snowsight during the prototyping phase. This is especially useful when teams want to compare prompt wording, model choice, and response quality on realistic sample inputs before operationalizing the solution with Cortex functions, SQL, Python, or an app. In Snowflake best practice, teams typically start with interactive validation in the Playground, then move to implementation once they understand which prompts and models perform best for the use case. By contrast, Snowpark Container Services is for more advanced custom hosting and application scenarios, Dynamic Tables are for data pipelines, and Streamlit is for app development rather than initial prompt/model exploration. Refer to Snowflake documentation on Cortex LLM Playground and Cortex AI for supported interactive prompt testing workflows in Snowsight.
- A. Correct.
Correct. Cortex LLM Playground is designed for interactive experimentation in Snowsight with prompts and supported large language models. It is useful early in development when teams want to compare model outputs and prompt variations on the same sample inputs before implementing the workflow in SQL, Python, or an application. This matches the scenario exactly: quick evaluation, no external tooling, and staying inside Snowflake.
- B. Incorrect.
Incorrect. Snowpark Container Services can be used for custom application and model-serving scenarios, but it is not the first tool for quickly comparing prompts and managed models in Snowflake. A common misconception is that model experimentation requires deploying your own endpoint first. Cortex LLM Playground is specifically intended to simplify early prompt and model evaluation without that infrastructure work.
- C. Incorrect.
Incorrect. Dynamic Tables are for incremental data transformation pipelines, not for interactive prompt engineering or side-by-side model exploration. Someone might choose this if they confuse operationalizing a pipeline with evaluating prompt quality. In this scenario, the team has not yet decided on a prompt or model and needs an experimentation interface first.
- D. Incorrect.
Incorrect. Streamlit in Snowflake is useful for building user-facing apps, but it is not required just to test prompts and compare model behavior. This distractor reflects the misconception that application development must start before prompt evaluation. Cortex LLM Playground is the lighter-weight starting point for interactive testing in Snowflake.