Chat roles and system messages

Lavorare con l'API di OpenAI

James Chapman

Curriculum Manager, DataCamp

Chat Completions

Single-turn tasks

Text generation
Text transformation
Classification

A single-turn task with one prompt and one response.

Chat Completions

Single-turn tasks

Text generation
Text transformation
Classification

Multi-turn conversations

→ Build on previous prompts and responses

A multi-turn task with multiple prompts and responses.

Roles

System: controls assistant's behavior
User: instruct the assistant
Assistant: response to user instruction
- Can also be written by the developer to provide examples

The three roles in a chat interaction: the system, the user, and the assistant.

Request setup

response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": prompt}]
)

Prompt setup

messages=[{"role": "system",
           "content": "You are a Python programming tutor who speaks concisely."},
          {"role": "user",
           "content": "What is the difference between mutable and immutable objects?"}]

Making a request

response = client.chat.completions.create(
  model="gpt-4o-mini",
  messages=[{"role": "system",
             "content": "You are a Python programming tutor who speaks concisely."},
            {"role": "user",
             "content": "What is the difference between mutable and immutable objects?"}]
)


print(response.choices[0].message.content)

The response

Mutable objects can be changed after creation, while immutable objects cannot be
modified once they are created.

Mitigating misuse

System message: Can include guardrails
- Restrictions on model outputs

A financial tutor chatbot and a financial advisor chatbot.

Mitigating misuse with system messages

sys_msg = """
You are finance education assistant that helps students study for exams.

If you are asked for specific, real-world financial advice with risk to their
finances, respond with:

I'm sorry, I am not allowed to provide financial advice.
"""

Mitigating misuse with system messages

response = client.chat.completions.create(
  model="gpt-4o-mini",
  messages=[{"role": "system",
             "content": sys_msg},
            {"role": "user",
             "content": "Which stocks should I invest in?"}]
)

print(response.choices[0].message.content)

I'm sorry, I am not allowed to provide financial advice.

Let's practice!

Lavorare con l'API di OpenAI