📄 中文摘要
在谈及“提示工程”时，许多人认为这是如何更好地向 AI 提问。实际上，这种看法是片面的。对于构建基于 API 的服务而言，提示实际上是在编程模型的行为。作者正在开发一个韩国算命应用和一个房地产分析平台，这两个项目都将基于大型语言模型（LLM）API。文中介绍了三种提示技术：零-shot、少-shot和链式思维。零-shot是指仅提供指令而不提供示例，适用于模型已知的任务；少-shot则是提供几个示例以帮助模型理解任务。通过这些技术，开发者可以更有效地利用 LLM API 来实现特定功能。
📄 English Summary
\"Ask ChatGPT nicely\" is amateur hour
When discussing 'prompt engineering,' many think it refers to how to ask AI better questions. This perspective is misleading. For those building services on top of APIs, prompting is about programming the model's behavior. The author is working on a Korean fortune-telling app and a real estate analysis platform, both utilizing LLM APIs. Three prompting techniques are highlighted: zero-shot, few-shot, and chain-of-thought. Zero-shot involves giving instructions without examples, suitable for tasks the model already knows; few-shot provides a few examples to help the model understand the task. These techniques enable developers to leverage LLM APIs more effectively for specific functionalities.
Powered by Cloudflare Workers + Payload CMS + Claude 3.5
数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等