0. Module Introduction: AI Novices vs. Advanced AI Users

Subject Areas: Human-Computer Interaction, AI Tool Usage, Information Retrieval

0.1 Topic Outline

• Novice AI users often use AI as a simple question and answer tool.
• Advanced AI users use AI for complex tasks with context, requiring models to think, use tools and iterate.
• This module focuses on Finding Information: how to determine whether to use pre-trained knowledge, web search, or deep research.

0.2 Key Knowledge Points

Core Requirements

• Core Differences in AI Advanced Users
  • Importance: ★★★★★
  • Difficulty: ★★★★☆☆☆☆
  • Definition: Advanced Users give the AI more difficult tasks, and give the model time, context, and evaluation criteria.
  • Example:
    • Novice asks “Does Taco Bell still have the Double Decker Taco?”
    • Advanced user asks “What are the tradeoffs for each car? Read everything and think hard before answering.”
  • Common test forms:
    • Given two prompts, determine which one is more appropriate for a complex task.
    • Rewrite the short prompt to include context, task objectives and output requirements.
  • An analogous association: novice users are like asking a librarian for just one word, advanced users are like asking a research assistant to complete a report with research questions, information, and scoring criteria.
  • Self-assessment question: Why is it often more valuable to “give an AI difficult tasks and ask it to think” than to just ask simple questions?
    • Answer: Because AI can consolidate information, compare scenarios, and form structured conclusions, thus saving manual reading and analysis time.

Understanding Mastery

• Provide High Quality Context
  • Importance: ★★★★★
  • Difficulty: ★★★☆☆☆☆
  • Conclusion: Short prompts are easy for AI to fill in the blanks and make guesses; by adding context such as project screenshots, voice notes, project documents, etc., the outputs are more relevant to the reality.
  • Example: When writing a self-review, it is more effective to provide materials of “what I have done” than just saying “Write a good self-review”.
  • Common test form: Adding the necessary context to the task.
• Neutral Questioning and Rating Scale
  • Importance: ★★★★☆
  • Difficulty: ★★★★★☆☆
  • Conclusion: Biased questions tend to get pandering answers; neutral questions with rubrics get more honest feedback.
  • Example:
    • Biased question: “I have a great business idea: mobile tie-dying. Critique it?”
    • 中性提问：“Analyze the following business idea objectively: mobile tie-dying. Use the rubric.”
  • Common Test Format: Identify inducing language in a prompt and change it to an objective evaluation of the prompt.

expand one’s understanding

• High-profile errors by AI do not represent the full extent of its capabilities
  • Importance: ★★★★☆☆☆☆
  • Difficulty: ★★★☆☆☆☆
  • CONCLUSION: The limitations of AI are important, but advanced users are able to bypass some of them and use it to write research papers, analyze personal data, and build websites.
  • Common test format: discuss the relationship between AI limitations and usage strategies.

0.3 Logical connections

The course begins with a comparison of AI novices and advanced users to illustrate that “the ability to ask questions” directly affects the value of AI. Then we move on to Finding Information, because information acquisition is the foundation of all complex AI workflows.

1. Pre-training knowledge: modeling learned public knowledge

Disciplinary Areas: Machine Learning, Natural Language Processing, Knowledge Representation

1.1 Outline of topics