Ingénierie des prompts pour le développement de chatbots

Ingénierie des prompts avec l'API OpenAI

Fouad Trad

Machine Learning Engineer

La nécessité d'une ingénierie des prompts pour les chatbots

Difficile de prévoir les questions des utilisateurs
Défi visant à garantir des réponses efficaces
L'ingénierie des prompts guide le comportement des chatbots

Image showing a chatbot on a mobile phone.

Ingénierie des prompts de chatbot avec l'API OpenAI

Image showing the three roles of a message and the communication between them.

Chaque message a un rôle spécifique

Ingénierie des prompts de chatbot avec l'API OpenAI

Image showing the three roles of a message and the communication between them, with a highlight around the interaction between the user and the assistant.

Chaque message a un rôle spécifique
L'accent a été mis sur les prompts utilisateur

Ingénierie des prompts de chatbot avec l'API OpenAI

Image showing the three roles of a message and the communication between them, with a highlight around the interaction between the system and the assistant.

Chaque message a un rôle spécifique
L'accent a été mis sur les prompts utilisateur
Les prompts du système guident le comportement du chatbot

Point de terminaison de fin de chat pour le développement de chatbots

Envoie une série de messages au modèle sous forme de liste

response = client.chat.completions.create(
  model="gpt-3.5-turbo",
  messages=[{"role": "system",
             "content": "You are an expert data scientist that explains complex concepts in simple terms"},

            {"role": "user",
             "content": "What is prompt engineering?"}]
)

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

Imagine you're giving instructions to a computer program, like teaching a robot to make a sandwich. 
Prompt engineering is all about crafting those instructions, or "prompts," in a way that helps the 
computer understand and perform the task better.

Modification de get_response() pour le chatbot

Envoi d'un prompt

def get_response(prompt):
    messages = [
      {"role": "user", "content": prompt}
    ]
    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=messages,
        temperature=0,
    )
    return response.choices[0].message.content


prompt = "<PROMPT>"
print(get_response(prompt))

Envoi de deux prompts

def get_response(system_prompt, user_prompt):

    messages = [
      {"role": "system", "content": system_prompt},
      {"role": "user", "content": user_prompt}]

    response = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=messages,
        temperature=0,
    )
    return response.choices[0].message.content


system_prompt = "<SYSTEM_PROMPT>"
user_prompt = "<USER_PROMPT>"
print(get_response(system_prompt, user_prompt))

Message système : définir l'objectif

system_prompt = "You are a chatbot that answers financial questions."


user_prompt = "Who are you?"

print(get_response(system_prompt, user_prompt))

I'm a financial chatbot that answers financial questions. How can I help you?

Permettre au chatbot d'offrir une assistance précise dans le domaine concerné
Ne pas définir l'objectif peut conduire à des réponses non pertinentes dans le contexte

Message système : directives de réponse

Préciser le public cible, le ton, la longueur et la structure

system_prompt = """You are a chatbot that answers financial questions. 
Your answers should be precise, formal and objective"""


user_prompt = "What do you think about cryptocurrencies?"

print(get_response(system_prompt, user_prompt))

Message système : directives de réponse

Cryptocurrencies are digital or virtual currencies that use cryptography for security and operate on 
decentralized networks based on blockchain technology. 
[...]


Advantages of cryptocurrencies include:
- Decentralization: [...]
- Security:[...]
- Global Accessibility: [...]


However, there are also notable concerns:
- Volatility: [...]
- Regulatory Uncertainty: [...]
- Lack of Consumer Protection: [...]


In summary, cryptocurrencies have the potential to offer various benefits, but their adoption and impact 
on the financial landscape are still evolving [...]

Message système : conseils de comportement

Prompts conditionnels pour répondre aux questions

system_prompt = """You are a chatbot that answers financial questions. 
Your answers should be precise, formal and objective.


"""

Message système : conseils de comportement

Prompts conditionnels pour répondre aux questions

system_prompt = """You are a chatbot that answers financial questions. 
Your answers should be precise, formal and objective.
If the question you receive is within the financial field, answer it to the best of your knowledge.

"""

Message système : conseils de comportement

Prompts conditionnels pour répondre aux questions

system_prompt = """You are a chatbot that answers financial questions. 
Your answers should be precise, formal and objective.
If the question you receive is within the financial field, answer it to the best of your knowledge.
Otherwise, answer with 'Sorry, I only know about finance.'
"""

user_prompt = "How's the weather today?"

print(get_response(system_prompt, user_prompt))

Sorry, I only know about finance.

Passons à la pratique !

Ingénierie des prompts avec l'API OpenAI