Researchers Reveal How to Enhance AI Response Quality

Carefully crafted prompts enhance AI response accuracy.

Carefully crafted contextual prompts improve the accuracy of AI model responses. This conclusion is drawn in a paper by the Shanghai Artificial Intelligence Laboratory.

Despite neural networks’ ability to understand natural language, they still require additional information and well-formulated queries to deliver quality results. For instance, if asked to “plan a trip,” an AI might suggest a luxury cruise without knowing about a limited budget.

Good questions help avoid “entropy”—confusion due to excessive uncertainty.

How to Formulate Quality Prompts

The paper suggests ways to enhance communication with artificial intelligence, based on prompt engineering.

Several tips include:

1. Start with the basics: who, what, why. Always include background to create context. Instead of the prompt “Write a poem,” try: “You are a romantic poet writing for my anniversary. The theme is eternal love. Keep the poem short and sweet.”
2. Layer information like a cake—from general to specific. Begin with the general, then add details. For a programming task: “I am a beginner programmer. First, explain the basics of Python. Then help debug this code [insert code]. Context: it’s for a simple game app.” This helps AI handle complex requests without overload.
3. Use tags and structure. Organize prompts with labels. Example: “Goal: plan a budget vacation. Constraints: $500. Suitable for family. Preferences: beach destination.” It’s akin to providing AI with a roadmap.
4. Include multimodal elements. If the request involves visual elements or previous chats, provide a description. Example: “Based on this image [description or link], suggest outfit options. Previous context: I prefer casual style.” For lengthy tasks, briefly outline the history.
5. Filter out noise. Include only the essentials in the prompt. If the AI “goes off track,” add clarification. For instance: “Ignore irrelevant topics—focus only on health benefits.”
6. Consider past mistakes. Think ahead, for example: “Last time you suggested X, but it didn’t work due to Y—adjust accordingly.”

Earlier in October, research from the University of Pennsylvania showed that large language models respond more accurately when addressed bluntly.