Puzzle Prompt Challenge
This repository explores the capabilities of Large Language Models (LLMs) in solving simple physical puzzles compared to human toddlers.
Background
I've noticed an intriguing situation: a young child can effortlessly solve simple puzzles, while current Large Language Models (LLMs) often struggle to provide clear, actionable, step-by-step solutions, even when given detailed prompts and photographs.
This repository documents a specific experiment to answer the question:
Can we effectively prompt-engineer LLMs into clearly solving simple physical puzzles?
Experiment
Input Image
Prompt
Here’s a picture of a child's puzzle. Please solve it in any way you see fit, but ensure the solution is obviously correct. I’d especially appreciate a generated image showing the puzzle completed, so it’s visually clear how all the pieces fit together.
LLM Response (Output Image)
How to Contribute
If you've tried similar experiments or have ideas about improving prompt effectiveness, contributions are welcome:
- Share your prompts and results.
- Suggest improvements or new experimental designs.
- Discuss findings and implications in GitHub issues.
License
This project is open source, licensed under the MIT License.