Databricks Generative AI Engineer Associate Question 132
Select 2You are working with a large language model (LLM) to generate responses for a customer service chatbot. The initial responses from the model are verbose and overly formal, but the desired output is concise and conversational. Which of the following prompt strategies would help adjust the model's responses to meet the desired output?
- A
Include an instruction in the prompt such as 'Respond concisely and in a conversational tone.'
- B
Fine-tune the LLM using a dataset of conversational and concise text examples.
- C
Specify a maximum word limit in the prompt, like 'Answer in 50 words or fewer.'
- D
Use a pre-trained conversational tone embedding and provide it as input to the model.
- E
Add an instruction in the prompt to include technical jargon where possible.
Show answer and explanation
Correct answers: A, C
Explanation
Adjusting an LLM's response from a baseline to a desired output can often be achieved through effective prompt design. Including specific instructions about tone and style and specifying limits, such as word count, are practical ways to achieve the desired behavior without modifying the model itself. Fine-tuning and embeddings are valid strategies but are outside the scope of prompt-based adjustments.
- A. Correct.
Including clear instructions in the prompt about how the LLM should respond, such as being concise and conversational, is an effective way to guide the model toward the desired output without requiring retraining.
- B. Incorrect.
Fine-tuning is a valid approach for customizing an LLM, but it is outside the scope of creating a prompt to adjust the model's response. This would require additional resources and is not relevant to the prompt-specific task.
- C. Correct.
Limiting the response length by specifying a word or character limit can help achieve concise outputs, provided the limit aligns with the desired conversational tone.
- D. Incorrect.
Using embeddings might be helpful in advanced use cases, but it is not directly related to prompt engineering or adjusting output via a prompt.
- E. Incorrect.
Adding technical jargon to the prompt would lead the LLM in the opposite direction of the desired output, which is conversational and concise.