A Framework for Collaborating a Large Language Model Tool in Brainstorming for Triggering Creative Thoughts
LLMs become better brainstorming partners through GPS (Goal, Prompt, Strategy) - a systematic prompt engineering toolkit.
LLMs become better brainstorming partners through GPS (Goal, Prompt, Strategy) - a systematic prompt engineering toolkit.
Original Problem 🔍:
Existing prompt engineering methods focus on accuracy, neglecting creativity enhancement in brainstorming with LLMs.
Solution in this Paper 🛠️:
• Introduces GPS framework: Goal, Prompt, Strategy
• Goal: Determines divergent/convergent thinking phase
• Prompt: Template with key elements (role, context, task, etc.)
• Strategy: Nine techniques to refine prompts (e.g., Chain-of-Thoughts, Unconventional Role)
• Adapts Torrance Tests of Creative Thinking for AI-generated idea evaluation
Key Insights from this Paper 💡:
• LLMs can enhance brainstorming creativity when guided by structured prompts
• Combining strategies can boost idea originality and elaboration
• Unconventional roles in prompts yield unexpected creative responses
• Framework seamlessly integrates into design practices
Results 📊:
• GPS template increased values across all creativity dimensions
• Strategies further enhanced flexibility, originality, and elaboration
• Some strategies boosted originality but reduced idea quantity
• Framework effective in generating both practical and imaginative ideas
The paper proposes several key strategies as part of the GPS framework:
Chain-of-Thoughts - Breaking down complex problems into steps
Unconventional Role - Having the LLM assume novel perspectives
Flipped Interaction - Making the LLM ask questions instead of answering
Analogy - Using examples to guide responses
Alternative Approaches - Exploring multiple solutions
Emphasis - Highlighting important aspects
Reflection - Having the LLM explain its reasoning
Environment Change - Altering the context
Self-Refinement - Iteratively improving responses
