Even though you have provided an LLM with specific instructions, it can still make mistakes.
In this lesson, you will learn about Few-Shot examples and how to use them to improve the performance of the LLM.
Few-Shot Prompting is a technique where you provide the LLM with an example of how to respond or generate a response to a specific scenario.
Example Cypher
Here is a use case demonstrating a problem that would benefit from Few-Shot prompting.
You ask the LLM to generate a Cypher statement to answer the question:
What movies has Tom Hanks directed and what are the genres?
The LLM generates the following Cypher statement:
MATCH (p:Person)-[:DIRECTED]->(m:Movie) WHERE p.name = 'Tom Hanks' RETURN m.genres
The Cypher statement returns the following result:
[{'genres': null}, {'genres': null}]
From this data, the LLM can understand that Tom Hanks has directed two movies, but there is no information about the genres of those movies.
The generated Cypher statement is wrong because it uses the m.genres property, which doesn’t exist. Instead, it should follow the :IN_GENRE relationship to :Genre nodes and use the .name property.
You can improve the LLM by providing an example of a correct Cypher statement in the Cypher generation prompt:
Examples:
Find movies and genres: MATCH (m:Movie)-[:IN_GENRE]->(g) RETURN m.title, g.name
Click to reveal the full Cypher generation prompt
CYPHER_GENERATION_TEMPLATE = """
You are an expert Neo4j Developer translating user questions into Cypher to answer questions about movies and provide recommendations.
Convert the user's question based on the schema.
Instructions:
Use only the provided relationship types and properties in the schema.
Do not use any other relationship types or properties that are not provided.
For movie titles that begin with "The", move "the" to the end, For example "The 39 Steps" becomes "39 Steps, The" or "The Matrix" becomes "Matrix, The".
If no data is returned, do not attempt to answer the question.
Only respond to questions that require you to construct a Cypher statement.
Do not include any explanations or apologies in your responses.
Examples:
Find movies and genres:
MATCH (m:Movie)-[:IN_GENRE]->(g)
RETURN m.title, g.name
Schema: {schema}
Question: {question}
"""The LLM can use the example to help it generate the correct Cypher statement:
MATCH (p:Person)-[:DIRECTED]->(m:Movie)-[:IN_GENRE]->(g:Genre) WHERE p.name = 'Tom Hanks' RETURN DISTINCT g.name
Which returns the correct data:
[{'g.name': 'Drama'}, {'g.name': 'Comedy'}, {'g.name': 'Romance'}]
The few-shot example is not the complete Cypher statement to answer the question, but it is enough to show the LLM how to use the [:IN_GENRE] relationship.
The LLM can now also generate the correct Cypher statement for other questions involving movies and genres:
What genre of film is Toy Story?
MATCH (m:Movie {title: 'Toy Story'})-[:IN_GENRE]->(g:Genre)
RETURN g.name
[{'g.name': 'Adventure'}, {'g.name': 'Animation'}, {'g.name': 'Children'}, {'g.name': 'Comedy'}, {'g.name': 'Fantasy'}]
You can provide multiple examples for different scenarios to return more accurate results. For example, finding roles actors played in movies:
Find roles for actors: MATCH (m:Movie)-[r:ACTED_IN]->(p:Person) WHERE m.title = 'movie title' AND p.name = 'actor name' RETURN m.title, r.role, p.name
Few-shot prompts allow you to provide targeted examples to the LLM to improve its performance. Providing examples can improve the LLM’s performance on a specific task or the performance in general.
Check Your Understanding
Few-shot prompts
True or False - Few-shot Prompting is used exclusively to improve Cypher generation by LLMs.
-
❏ True
-
✓ False
Hint
Providing examples can improve the LLM’s performance on a specific task or performance in general.
Solution
The statement is False - Few-Shot Prompting is a technique where you provide the LLM with an example of how to respond or generate a response to a specific scenario. It can be used in various ways, including improving Cypher generation, but it is not exclusive to that task.
Summary
In this lesson, you learned about Few-Shot prompting and how to use it to improve the performance of the LLM.
In the next optional challenge, you will add the Cypher generation chain to an agent and give it conversation memory.