Generating a Related Work Section with AI

This exercise gives you hands-on experience using Large Language Models (LLMs) to accelerate writing a "Related Work" section. You will explore two perspectives: that of an **expert** and a **non-expert**. The goal is to understand the process, learn to craft effective prompts, and critically evaluate the AI's output.

1. Generate Related Work

As both an expert and non-expert, use an LLM to generate a "Related Work" section.

2. Propose New Ideas

After generating the related work, use the LLM to identify gaps and propose new research directions.

3. Document Everything

Keep a Google Doc to record your prompts, responses, and reflections throughout the process.

4. Expert Evaluation

After swapping back, the expert evaluates the non-expert's generated content and ideas.

Before You Begin: Setup for Success

  • ①

    Start a Fresh Chat Session

    To prevent context from previous conversations from influencing your results, start a new, temporary, or incognito chat session in your LLM for this exercise.

  • ②

    Keep Your Notes Open

    Have your Google Doc open the entire time. Your goal is to document your process **as it happens**. Don't wait until the end. For each attempt, record:

    • The exact prompt you used.
    • A brief summary of the AI's output.
    • Your own thoughts on what worked or didn't work.

Part 1 & 2: Strategies & Ideation

You as the Expert

Key Strategy: Provide Scaffolding and Constraints

When you're familiar with the research area, you can guide the LLM with much greater precision. You know the key players, seminal papers, and schools of thought. Use this knowledge to your advantage.

Part 3: Expert Evaluation

After completing the exercises and swapping back, the "expert" should review the work generated by the "non-expert." The goal is not just to grade, but to provide constructive feedback on the AI-assisted process from a place of deep domain knowledge.

Evaluation Checklist

  • ✔

    Accuracy of Related Work

    Did the non-expert, with the AI's help, correctly identify the key competing and complementary research? Was the summary of the field accurate?

  • ✔

    Plausibility of New Ideas

    Are the proposed new ideas interesting? Are they grounded in a reasonable understanding of the field, or do they stem from a misunderstanding? Could any of the "naive" ideas actually be innovative?

  • ✔

    Identification of Gaps

    Did the non-expert's process successfully identify a legitimate gap in the literature that your paper addresses? Did they identify other gaps you might not have considered?

General Prompting Tips & Tricks

🎭 Use Personas

Tell the LLM who it should be. E.g., "Act as an expert in computational linguistics..."

🎯 Be Clear and Specific

Vague prompts lead to vague answers. Provide concrete details and constraints.

📝 Provide Examples

If you want a specific style or format, give the LLM a short example to follow.

🔄 Don't Be Afraid to Start Over

If a line of prompting isn't working, open a new chat and try a different approach.

✅ Fact-Check Everything

LLMs can "hallucinate" citations and misrepresent findings. Always verify.

Reflection Questions