نوع مقاله : مقاله پژوهشی
عنوان مقاله English
Introduction: Large language models have become effective tools in specialized research; however, the quality and accuracy of their outputs depend on the clarity, relevance, level of detail and information, and structure of the prompt. This study examines the fundamental principles of designing and writing artificial intelligence prompts, with an emphasis on their application in halal studies.
Methods: This descriptive-analytical study was conducted using a library-based method and draws on the author’s several years of experience in designing, evaluating, and optimizing specialized prompts. The proposed principles were developed with a focus on issues such as jurisprudential rulings, standards, and specialized terminology. They were also tested and validated in three language models: ChatGPT, Claude, and Gemini.
Results: The extracted principles include clarity and explicitness; avoiding ambiguity; using action-oriented and imperative verbs; separating instructions from input; distinguishing different prompt components through XML tags; providing contextual information and examples; specifying the output format; prioritizing quantification; maintaining an appropriate level of detail; conciseness; using specialized terminology; posing direct and neutral questions; proceeding step by step; using standardized formats; testing diverse prompt structures; allowing the models to express uncertainty to mitigate hallucinations; repeatedly reviewing and optimizing prompts; and managing attention in lengthy conversations.
Conclusion: Applying these principles in halal studies can improve the quality, efficiency, and structural coherence of model outputs, increase their accuracy and alignment with research needs, and purposefully guide the models’ probabilistic space. These principles can also facilitate and accelerate research.
کلیدواژهها English