AI Tips for Engineering? All Programmers Really Need Are These Three Tips!
Revealing the efficient skills of AI prompt engineering, three secrets that programmers must learn!
Core content:
1. The three most effective prompting skills in the eyes of programmers: context learning, thinking chain, and structured output
2. Using professional language to improve the effect of AI prompts, the ability to transfer knowledge in different fields is limited
3. Simple and clear prompts are more effective, avoiding irrelevant information that reduces the cognitive efficiency of the model
AI Tips for Engineering? All Programmers Really Need Are These Three Tips!
This article comes from an interesting discussion thread: https://news.ycombinator.com/item?id=44182188
Have you ever doubted the so-called "prompt engineering"? Some people think it is too mysterious, while others call it a "new outlet". So, what does prompt engineering look like in the eyes of programmers?
Today, let’s put aside the gimmicks and look at what tips and tricks actually work.
1. Three Tips from the Programmer’s Perspective
In actual development and exploration, there are actually three most effective prompting techniques:
- In-Context Learning
Provide examples or context, such as “one-shot” or “few-shot.” AI performance improves dramatically when it has context. - Chain of Thought
Let AI think step by step (for example, prompt “Please think step by step”) to improve its ability to handle complex tasks. - Structured Output
Explicitly specifying the output format, such as JSON, can help avoid confusing text replies.
In addition to the above three, some people may also mention "Role Prompting" or "Retrieval Augmented Generation" (RAG), but in the final analysis, the core of these methods is to clearly and accurately describe the requirements - context is crucial.
2. “Professional language” improves prompt effect
Programmers’ experience shows that LLM’s ability to transfer knowledge in different fields is still limited, and the right “language” is particularly critical. For example:
When you ask AI questions using technical terms, the quality of the answers you get improves significantly. A doctor who describes symptoms using medical abbreviations will get an accurate diagnosis, whereas a more general description will often yield little results.
Therefore, if you want better results, you have to use the professional language of the target field to "tune" AI.
3. The simpler the prompts, the more effective they are?
Sometimes, simplicity means efficiency. Concise and clear prompts often work better. On the contrary, a large amount of irrelevant information will only reduce the cognitive efficiency of the model and even bring additional costs.
The key to prompt engineering is not to create "profound" prompts, but to clearly and unambiguously communicate the requirements. It is more practical and economical to give AI appropriate context and target assumptions, examine the results, and iterate and improve.
4. The future: Will prompt engineering become a programming language?
Perhaps in the next few years, a dedicated prompt language will emerge, and "prompt engineering" may really become a new "programming paradigm." Just like SQL, we will be able to accurately "control" AI through a specific language.
But at this stage, it is more important to master the basics first: clear context, effective examples, and clear output.
5. How should AI be used?
Instead of worrying about whether the prompt is "engineering" or not, it is better to regard AI as your assistant rather than "magic". You can really use AI well by doing the following three things:
Clear expression of needs Appropriate professional language Concise context
From a programmer's perspective, if you master the above three points, you can also become a true "efficient user of AI."
