吴恩达 ChatGPT Prompt Engineering for Developers 系列课程笔记--07 Expanding

Isawany

已于 2023-06-07 22:47:14 修改

阅读量756

点赞数

分类专栏：吴恩达ChatGPT系列课程学习文章标签： chatgpt OpenAI 吴恩达语言模型课程学习

于 2023-06-07 20:42:11 首次发布

本文链接：https://blog.csdn.net/weixin_38124427/article/details/131037345

版权

吴恩达ChatGPT系列课程学习专栏收录该内容

9 篇文章 3 订阅

订阅专栏

07 Expanding

本节示例如何用ChatGPT生成一封电子邮件的回复。

1) 定制化情绪

给定客户评论，我们根据评论内容和情绪产生定制的回复。下面是给定情感（positive/negative），让ChatGPT产生相应回复的prompt。

"""
You are a customer service AI assistant.
Your task is to send an email reply to a valued customer.
Given the customer email delimited by ```, \
Generate a reply to thank the customer for their review.
If the sentiment is positive or neutral, thank them for \
their review.
If the sentiment is negative, apologize and suggest that \
they can reach out to customer service. 
Make sure to use specific details from the review.
Write in a concise and professional tone.
Sign the email as `AI customer agent`.
Customer review: ```{review}```
Review sentiment: {sentiment}
"""

虽然prompt很长，但关键任务其实就是Generate a reply to thank the customer for their review.前两句是情景假定，可以让ChatGPT明确环境，下面一句指定由三引号分隔原邮件防止恶意注入。后面几句话则是对输出的要求。
最后指定回复的情感。情感为negative时ChatGPT的回复为

Dear valued customer,

Thank you for taking the time to leave a review about our product. We are sorry to hear that you experienced a price increase and that the quality of the product did not meet your expectations. We apologize for any inconvenience this may have caused you.

情感指定为positive时ChatGPT的回复为

Dear Valued Customer,

Thank you for taking the time to leave a review of our 17 piece system. We are thrilled to hear that you were able to take advantage of our seasonal sale in November and that you found the product to be useful in your daily routine. We appreciate your detailed feedback on the product and your tips for making smoothies.

当然我们也可以不指定情感让ChatGPT自己去识别。

2) temperature

temperature是语言模型的一个重要参数，用于控制生成回复的随机性。比如下图中，假设输入"my favorite food is"给ChatGPT，模型预测下一个token可能是"pizza", “sushi”, “tacos”, …。temperature为0时，模型会倾向于选择概率更高的token输出，三次的结果可能都是pizza（53%的概率），当设置temperature=0.3时，模型会增加输出的随机性，可能会选择概率更小的suchi，当然还会可能选择pizza。总之temperature越大，输出的不确定性越大。

笔者另外做了个简单的小测试，设置prompt = f""" Finish the sentence delimited by ```by adding one word ```Dog likes eating``` """，可以看到temperature=2时，ChatGPT给出各种可能的答案:

temperature= 0
response 1: bones.
response 2: bones.
response 3: bones.
temperature= 0.5
response 1: bones.
response 2: bones.
response 3: bones.
temperature= 1.0
response 1: bones.
response 2: bones.
response 3: bones.
temperature= 2.0
response 1: feet.
response 2: Bones everyday.
response 3: bones.

一般来说，为了使生成的结果可信赖、可预测，我们设置temperature=0（每次得到相同的结果）。当编写故事时如果想更有意思，可以设置更高的temperat