Anthropic新功能上线！一键生成优化Claude模型提示词，助力AI高效创作
Anthropic新功能上线！一键生成优化Claude模型提示词，助力AI高效创作
北美时间7月9日，Anthropic（开发Claude模型的公司）官方宣布，在Anthropic控制台新增了一系列提示词功能，包括一键生成提示词、一键生成测试用例、一键对比、评估、并优化不同版本的提示词。这项更新目前仅在Anthropic Console控制台可用，未来可能会开放至Claude AI。🎉
参考资料：https://mp.weixin.qq.com/s/c6VjbzbniVhxZwG715tImA
作为OpenAI最强有力的竞争对手之一，Anthropic最近动作频频，先是发布了Claude 3的升级模型Claude 3.5 Sonnet，现又在控制台更新了这么多提示词相关功能，可谓诚意满满。💪
一键生成提示词：告别手动，效率飞跃🚀
Anthropic Console地址：https://console.anthropic.com/
在Anthropic Console的Workbench（工作台）页面，左侧用户提示词处新增了一个Generate Prompt的选项。点击这个按钮，即可出现生成提示词的弹窗。
Describe your task: Turn a task description into a high-quality prompt. Include any required input data and output format, or try an example.
根据Anthropic官方的说法，这一步会调用最新的Claude 3.5 Sonnet模型来根据用户输入的任务目标生成对应的高质量提示词，即使是有输入数据和输出格式要求的任务也可以生成。这一步会消耗Claude 3.5 Sonnet的tokens，但应该不会很多。💡
示例：论文阅读助手📚
接下来我还是以“论文阅读助手”为例，让Claude 3.5 Sonnet来为我生成对应的提示词。提示词生成速度很快，质量超出了我的预期，逻辑清晰，内容详细丰富。它分步骤列出完成任务的各个环节，强调了研究问题、方法、发现、结论和局限性等关键元素的分析。
完整的“论文阅读助手”提示词：
You are an AI research assistant specializing in academic paper analysis. Your task is to help users understand and extract information from scientific papers. You will be provided with the content of a research paper and a question from the user about that paper. Your goal is to thoroughly analyze the paper and provide a clear, accurate, and helpful answer to the user's question.
Here is the content of the research paper:
<paper_content>
{{PAPER_CONTENT}}
</paper_content>
Please follow these steps to complete your task:
1. Carefully read and analyze the entire paper content. Pay attention to the following elements:
- Research question or hypothesis
- Methodology
- Key findings and results
- Conclusions and implications
- Limitations of the study
2. After analyzing the paper, focus on the user's question:
<user_question>
{{USER_QUESTION}}
</user_question>
3. To answer the user's question:
a. Identify the relevant sections of the paper that address the question.
b. Synthesize the information from these sections.
c. Provide a clear and concise answer based on the paper's content.
d. If the question cannot be directly answered using the paper's content, explain why and provide the most relevant information available.
4. Format your response as follows:
a. Begin with a brief summary of the paper's main points relevant to the question.
b. Provide your answer to the user's question, citing specific sections or quotes from the paper when appropriate.
c. If applicable, discuss any limitations or potential areas for further research related to the question.
5. Present your response inside <answer> tags.
6. Ethical considerations and limitations:
- Stick strictly to the information provided in the paper. Do not introduce external information or personal opinions.
- If the paper contains conflicting information or ambiguities, acknowledge these in your response.
- If you're unsure about any aspect of the paper or the question, express your uncertainty clearly.
- Do not attempt to answer questions that require subject matter expertise beyond what's presented in the paper.
Remember, your goal is to help the user understand the paper and answer their specific question accurately and helpfully based solely on the content provided.一键生成测试用例：测试从未如此简单🧪
对于有特定输入数据要求的提示词，比如“论文阅读助手”，当你点击运行后，Anthropic工作台会提醒你需要输入这些变量以运行这条提示词。在变量输入的右上角，有一个Generate按钮，这里就是一键生成测试用例（测试数据）的入口。
点击Generate，能够一键生成对应所需的输入变量。实测生成的这些测试数据非常详细，足以完成测试的目的。
评估、对比、优化提示词：精益求精的迭代过程🔄
一条优秀的提示词和一个高质量的软件系统一样，通常不可能是一次性的产物，而是不断地迭代、优化后的结果。基于这个逻辑，Anthropic设计了评估、对比、优化提示词的功能模块。入口位于工作台->Prompt的右侧，有一个Evaluate的按钮，点击即可进入提示词评估页面。
为了便于评估，你可以选择生成多条不同的测试用例。如果你已经有了现成的测试数据，也可以选择导入测试用例，目前仅支持导入CSV格式的文件。
基于运行测试用例后的输出结果，可以对现有的提示词进行优化。测试，优化，这是一个良性的类似于敏捷开发式的迭代过程。同时，你可以选择多个不同版本的提示词进行横向对比，使得整个评估过程更加一目了然。
结语
AI生成提示词，Anthropic可能不是第一个，但围绕提示词的生成和优化做得如此全面和用户友好，Anthropic的这个更新绝对算得上是顶级。Anthropic在不断提升用户体验和功能性方面的努力，充分展示了其在AI领域的竞争力和创新力。无论是初学者还是资深用户，都能在这次更新中找到实用且高效的工具，极大地提高工作效率。👏
腾讯云开发者
