

大家好,这里是程序员晚枫,正在all in AI编程实战。
第14讲:Prompt工程——写出高质量的AI提示词
为什么Prompt这么重要?
同样的AI模型,不同的提示词,效果天差地别。
1、Prompt的5个核心要素
| 要素 | 说明 | 示例 |
|---|
| 角色 | 让AI扮演谁 | 你是一位Python教师 |
| 任务 | 要AI做什么 | 写一篇入门教程 |
| 背景 | 上下文信息 | 面向零基础小白 |
| 格式 | 输出要求 | Markdown格式,含代码块 |
| 约束 | 限制条件 | 字数800字,不用专业术语 |
2、角色设定技巧
1 2 3 4 5 6 7
| system_prompt = """你是一位经验丰富的Python编程老师,名叫"程序员晚枫"。
你的教学风格: - 通俗易懂,用生活比喻解释编程概念 - 循序渐进,从简单到复杂 - 多用代码示例,少用理论 """
|
3、Few-shot:给AI举例
1 2 3 4 5 6 7 8
| messages = [ {"role": "system", "content": "你是Python老师"}, {"role": "user", "content": "什么是变量?"}, {"role": "assistant", "content": "变量就像一个盒子,可以装东西。"}, {"role": "user", "content": "什么是函数?"}, {"role": "assistant", "content": "函数就像一个菜谱,写好步骤,需要时直接调用。"}, {"role": "user", "content": "什么是循环?"} ]
|
4、结构化输出
1 2 3 4 5 6 7 8 9 10 11 12 13 14
| prompt = """请分析以下文本的情感,以JSON格式输出: { "sentiment": "positive/negative/neutral", "confidence": 0.0-1.0, "keywords": ["关键词1", "关键词2"] }
文本:这个产品真的太棒了,强烈推荐!"""
response = client.chat.completions.create( model="deepseek-chat", messages=[{"role": "user", "content": prompt}], temperature=0 )
|
5、思维链(Chain of Thought)
1 2 3 4 5 6 7 8
| prompt = """请一步一步思考以下问题:
一个商店,进货价50元,售价80元,每天卖出30件。
请按以下步骤计算: 1. 每天毛利润 2. 每天净利润 """
|
6、实战:构建一个文本分析工具
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
| from openai import OpenAI import json
client = OpenAI(api_key="你的Key", base_url="https://api.deepseek.com")
def analyze_text(text): prompt = f"""分析以下文本,以JSON格式返回: {{ "sentiment": "positive/negative/neutral", "confidence": 0.0到1.0, "keywords": ["关键词1", "关键词2"], "summary": "一句话总结" }} 文本:{text}"""
response = client.chat.completions.create( model="deepseek-chat", messages=[{"role": "user", "content": prompt}], temperature=0 ) return json.loads(response.choices[0].message.content)
result = analyze_text("Python太好用了,我学了一周就能写自动化脚本!") print(f"情感:{result['sentiment']}") print(f"关键词:{result['keywords']}")
|
7、常见问题
Q:AI总是输出格式不对?
A:在prompt里明确写出格式示例,temperature设为0。
Q:回答太长/太短?
A:明确指定字数或条数。
下讲预告
学会了Prompt工程,下一讲我们学 RAG入门——让AI读取你的文档。
敬请期待!
程序员晚枫专注AI编程培训,小白看完他和图灵社区合作的教程《30讲 · AI编程训练营》就能上手做AI项目。
前3讲可以试听,试听链接:https://www.bilibili.com/cheese/play/ss982042944