How to Use Large Language Models Efficiently: A Practical Guide from the llm-queries Project

This article analyzes the GitHub open-source project llm-queries (maintained by djeada, original link: https://github.com/djeada/llm-queries, updated on 2026-06-16), sharing core techniques for efficient collaboration with GPT and other large language models, covering key practices such as prompt engineering, context management, and output optimization, helping users move from "being able to use" LLMs to "using them well".

The llm-queries project was born from the observation that most people interact with LLMs inefficiently, often resulting in unsatisfactory outputs due to vague prompts. Its core goal is to integrate scattered best practices into a structured knowledge base, helping users significantly improve output quality and work efficiency in scenarios such as code writing, document creation, data analysis, or creative writing.

The project emphasizes three basic principles:

• Clarity and Specificity: Vague requests tend to result in vague answers; you need to specify output format, level of detail, target audience, etc.
• Provide Context: The model has no knowledge of the current situation; you need to provide background information such as identity, goals, known conditions, etc.
• Decompose Complex Tasks: Split multi-step tasks into simple subtasks to increase success rate and result verifiability.

Advanced techniques include:

• Role-Playing: Assign specific roles (e.g., senior architect, primary school teacher) to activate domain knowledge and expression styles.
• Example-Driven (Few-shot Prompting): Embed input-output examples to guide the model in understanding expected patterns, suitable for scenarios like data conversion and code generation.
• Chain-of-Thought Prompting: Require the model to show its reasoning process to enhance accuracy and transparency for complex problems such as mathematics and logic.

The project provides cases for multiple scenarios:

• Code-Assisted Development: Request code reviews, generate unit tests, explain algorithms, etc.; you need to provide code context and framework version.
• Document and Content Creation: First generate an outline then expand the content; achieve style adjustments, tone conversion, and multilingual translation.
• Data Analysis and Interpretation: Describe data format and requirements; assist in data cleaning, statistical analysis, and result interpretation.

Pitfalls to avoid:

• Over-Reliance and Fact-Checking: The model may generate incorrect information; key content needs independent verification.
• Over-Engineering of Prompts: Avoid complex prompts; start with simple ones and gradually iterate and optimize.
• Ignoring Model Limitations: Understand model boundaries such as context length and knowledge cutoff date; set reasonable expectations.

Core Points: Clarify requirements, provide context, decompose tasks, make good use of roles and examples, and maintain critical thinking. Action Recommendations: Practice basic skills first, gradually try advanced patterns, develop a personalized prompt style, and reflect continuously. These skills are important competencies in the AI era, and the llm-queries project provides a valuable starting point for learning.