【大模型应用开发极简入门】构建新闻稿生成器：提示词的使用与基于事实的提示词

GPT-4和ChatGPT等LLM专用于生成文本。我们可以使用GPT-4和ChatGPT在各种场景中生成文本，举例如下。

• 电子邮件
• 合同或正式文档
• 创意写作
• 逐步行动计划
• 头脑风暴
• 广告
• 职位描述

对于本项目，我们将创建一个工具，它可以根据一系列事实生成新闻稿。我们可以根据目标媒体和受众选择新闻稿的篇幅、语调和风格。

一. 提示词怎么写

这里主要描述prompt（提示词）的构建逻辑，因为大模型可以根据prompt的规定生成符合要求的文档。

1. 给AI模型分配一个角色，并尽可能精确地描述任务。如下给AI模型分配的角色是记者助手：

prompt_role = "You are an assistant for journalists. \Your task is to write articles, based on the FACTS that are \given to you. \You should respect the instructions: the TONE, the LENGTH, \and the STYLE"

2. 其他规定

• prompt_role：角色的描述，以便大模型能够按照角色回答
• FACTS：基于给定的事实数据来回答
• TONE：回答风格：这里是informal
• LENGTH：回答的单词数
• STYLE：生成的文本格式：这里是blogpost

# 拼装messages，规定了prompt的格式：  
# prompt_role：角色的描述，以便大模型能够按照角色回答  
# FACTS：基于给定的事实数据来回答  
# TONE：回答风格：这里是informal  
# LENGTH：回答的单词数  
# STYLE：生成的文本格式：这里是blogpost  
def assist_journalist(  facts: List[str], tone: str, length_words: int, style: str  
):  facts = ", ".join(facts)  prompt = f"{prompt_role} \  FACTS: {facts} \  TONE: {tone} \  LENGTH: {length_words} words \  STYLE: {style}"  return ask_chatgpt([{"role": "user", "content": prompt}])

 

二. 完整代码

import os  import openai  
from typing import List  openai.api_key = os.getenv('OPENAI_API_KEY')  # 调用openai api  
def ask_chatgpt(messages):  response = openai.ChatCompletion.create(  model="gpt-3.5-turbo", messages=messages  )  return response["choices"][0]["message"]["content"]  # prompt_role描述  
prompt_role = "You are an assistant for journalists. \  Your task is to write articles, based on the FACTS that are \  given to you. \  You should respect the instructions: the TONE, the LENGTH, \  and the STYLE"  # 拼装messages，规定了prompt的格式：  
# prompt_role：角色的描述，以便大模型能够按照角色回答  
# FACTS：基于给定的事实数据来回答  
# TONE：回答风格：这里是informal  
# LENGTH：回答的单词数  
# STYLE：生成的文本格式：这里是blogpost  
def assist_journalist(  facts: List[str], tone: str, length_words: int, style: str  
):  facts = ", ".join(facts)  prompt = f"{prompt_role} \  FACTS: {facts} \  TONE: {tone} \  LENGTH: {length_words} words \  STYLE: {style}"  return ask_chatgpt([{"role": "user", "content": prompt}])  print(  assist_journalist(  ["The sky is blue", "The grass is green"], "informal", \  100, "blogpost"  )  
)

 

输出如下

"Hey, everyone! Did you know that the sky is blue and the grass is green?
I mean, it's something we see every day and probably take for granted,
but it's still pretty amazing if you think about it! The sky appears
blue to us because of something called Rayleigh scattering – basically,
the molecules in the Earth's atmosphere scatter sunlight in all different
directions. Blue light has a shorter wavelength, so it gets scattered
more than the other colors in the spectrum. That's why the sky looks
blue most of the time! As for the grass being green... that's due to
chlorophyll, the pigment that helps plants capture sunlight to make
their food. Chlorophyll absorbs red and blue light, but reflects
green light, which is why we see plants as green.It's pretty cool how science explains these things we take for granted,
don't you think? Next time you're outside, take a moment to appreciate
the color palette around you!"

 

三. 基于事实的prompt

通过明确facts数据，让GPT基于事实来回答。

print(assist_journalist(# 这里让facts=["A book on ChatGPT has been published last week","The title is Developing Apps with GPT-4 and ChatGPT","The publisher is O'Reilly.",],tone="excited",length_words=50,style="news flash",)
)

结果如下：

Exciting news for tech enthusiasts! O'Reilly has just published a
new book on ChatGPT called "Developing Apps with GPT-4 and ChatGPT".
Get ready to delve into the world of artificial intelligence and learn
how to develop apps using the latest technology. Don't miss out on this
opportunity to sharpen your skills!