Clarify with Questions
By encouraging our model to generate and answer clarifying questions before tackling the main query, we can obtain more accurate and comprehensive responses. This is known as Self-Ask 1.
We can implement this using instructor as seen below.
import instructor
from openai import OpenAI
from pydantic import BaseModel, Field
client = instructor.from_openai(OpenAI())
class FollowupQuestion(BaseModel):
question: str = Field(
description="Question to be answered",
)
answer: str
class SelfAskResponse(BaseModel):
follow_up_questions: list[FollowupQuestion] = Field(
description="""A list of question and
answer pairs that are required to be
answered in order to answer the original
question.""",
default_factory=list,
)
answer:str
def generate_questions_and_response(query):
return client.chat.completions.create(
model="gpt-4o",
response_model=SelfAskResponse,
messages=[
{
"role": "system",
"content": """You are a helpful assistant
which is able to accurately answer user
queries. Generate a list of follow up
questions and responses before answering
the user's question below.""",
},
{
"role": "user",
"content": query,
},
],
)
if __name__ == "__main__":
query = """Who was president of the U.S.
when superconductivity was discovered?"""
response = generate_questions_and_response(query)
print(response.model_dump_json(indent=2))
"""
{
"follow_up_questions": [
{
"question": "When was superconductivity discovered?",
"answer": "Superconductivity was discovered in 1911 by
Heike Kamerlingh Onnes."
},
{
"question": "Who was president of the U.S. in 1911?",
"answer": "The president of the U.S. in 1911 was William
Howard Taft."
}
],
"answer": "The president of the U.S. when superconductivity was
discovered in 1911 was William Howard Taft."
}
"""
References¶
1: Measuring and Narrowing the Compositionality Gap in Language Models