Скачать или смотреть Prompt Engineering Techniques for Software Development: Controlled Versus Uncontrolled Prompts!

Controlled versus Uncontrolled Generative Prompts
Uncontrolled Prompt
An uncontrolled prompt is a generative prompt that requests a solution without specific constraints, leaving the AI to decide how to approach the task. It is undescriptive and open-ended.

High degree of freedom
Best for simple tasks that do not require design coordination with an architecture plan
Less programmatic
More opportunity for hallucination
Benefits:
Minimal effort – You may get lucky and receive a good result without much work.
Flexibility – You don’t need to already have a solution in mind.
Speed – Quick to execute, as no prior thought is required to shape the prompt.
Drawbacks:
Unpredictability – Since the prompt is open-ended, there is a heavy reliance on luck.
Lack of direction – The result will not necessarily follow a coherent organizational scheme imposed by you on it, and by extension, your entire project.
Risk – Introducing unsystematic, uncontrolled output into a designed system can be risky.
Inconsistent output – A codebase or project built on uncontrolled prompts may lack consistency and organization.
Controlled Prompt
A controlled prompt is one that requests a solution according to specific constraints and specifications. It is detailed and descriptive to keep the AI within well-defined boundaries

Low degrees of freedom
Necessary for more complex projects that require consistency in overall design
More programmatic
Low opportunity for hallucination (easily detected)
Benefits:
Intentionality – You impose your scheme, structure, and logic on the model’s output.
Clear evaluation – You can assess results against self-defined pass/fail criteria.
Faster development (when well-prepared) – If you already understand the solution you need, this approach will get you there faster by outsourcing implementation to the AI.
Engineering discipline – By enforcing an organized and systematic approach, controlled prompts lead to more consistent, high-quality output.
Essentially, you are handling the pseudo-code—the high-level logic—and leaving the lower-level implementation details to the AI. In fact, you can even specify implementation requirements when necessary.

Drawbacks:
Prerequisite knowledge – You need to have a clear understanding of the problem and the desired solution in order to provide detailed constraints.
More effort – It requires more upfront work to design a controlled prompt, though this can ultimately save time in the long run.
Potential frustration – If the AI fails to adhere to your specifications, you may need to intervene directly to ensure the solution is properly implemented.
“If you want something done right, do it yourself.”
Summary
For controlled prompts, it is essential to emphasize control language over mere descriptiveness. While descriptive language communicates what you want, control language highlights that you are imposing a structured plan on the AI’s output. For instance:

Give me X under Y and Z conditions.

This approach ensures the AI operates within the boundaries of your plan, minimizing unpredictability and enhancing alignment with your project’s goals.