用于 Web 开发的提示工程 — SitePoint

Future of Code Writing: In coming years, there will be a shift in code writing. Understanding large learning models (LLM's) and interacting with them through prompts will be crucial. Many are already doing this as it's valuable to generate code on demand.
- Code Generation Experiment: The author has been experimenting with code generation for 2 years and has seen a big leap in the past 6 months. This is expected to become more common, but fundamental concepts for interacting with AI prompts will always be important.
Prompt Engineering: This involves sharing approaches and techniques for generating code. The quality of the input directly affects the output. For example, when asking an AI to write code, be super clear, mention relevant details like WordPress, use built-in functions, and provide specific hooks. These aspects narrow down the model's focus and result in a higher quality output.
Code Organization & Reuse: To write effective AI prompts, one needs to have a good understanding of systems and tools. Well-crafted prompts make use of the current codebase in the simplest ways. AI-generated code can get out of hand if functions are independent. Keeping code clean and organized is essential, whether by humans or AI.
Good Prompt vs Bad Prompt: Vague prompts result in partially complete answers or instructions on how to fill in gaps. For example, a prompt to make an API request for latest Reddit posts on a website without mentioning WordPress or providing enough context will likely lead to less useful code. A good prompt, on the other hand, provides detailed instructions and relevant context, resulting in significantly better code.
Advanced Example: An example shows how to create a complex prompt to generate custom WP CLI commands for generating SEO reports, creating posts from markdown, and finding broken links. This requires specific details and knowledge of WordPress programming.
The Cost of Easy: AI code generation makes programming faster and accessible, but it also makes it possible for people to write code without understanding how it works. This can lead to poorly architected code that may cause problems in the long run.
Conclusion: Documenting prompts is time-consuming but worth it. Writing quality prompts that result in precise code will be a critical skill for developers. Being fluent in the programming language and knowing the codebase inside out will still be important. AI code generation will continue to revolutionize website and application building.