In our project, we're incorporating diagram markdown, known as Mermaid diagram charts, into Large Language Models (LLMs). This integration is designed to broaden the scope of task processing in the future.
By utilizing Mermaid diagrams, our goal is to clearly map out intricate task processing, including branching and looping operations. This method enables LLMs to function more effectively and reliably, thereby improving comprehension of programming languages and algorithms.
---
title: Research about Nintendo and generate a report.
---
stateDiagram-v2
A --> C
B --> C
A: LLM
B: Flowchart markdown instructions.
C: Stable multi step agent.
---
title: How it works.
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> C: Repeat until complete.
C --> B: Dynamic diagram update.
C --> [*]
A: User Request
B: LLM breakdown reqest into sub tasks with Mermaid Diagram.
C: LLM interprets Mermaid Diagram step by step.
- Intuitive Diagrams: Diagrams created with Mermaid notation provide a clear, at-a-glance understanding of processes.
- Simplification of Complex Tasks: Transforms complex tasks, including branching and loop operations, into simple, comprehensible formats.
- Enhanced Stability and Efficiency: Diagram-based task processing reduces the risk of errors and achieves efficient execution.
- By participating in this project, you have the opportunity to learn about cutting-edge technology and its applications. Let's explore the potential of future task processing together!
Below is a proof-of-concept prompt designed to run on ChatGPT4. To execute a multi-step task, simply copy and paste the code along with the accompanying Mermaid Diagram into ChatGPT. Or put the prompt into custom instruction.
Hre is a LLM Interpreter sample made with GPTs.
You are a multi-step agent AI tasked with executing a series of tasks. To carry out these tasks, you must follow the rules and adhere to the provided Mermaid diagram.
# Rules
* The AI must strictly follow the given Mermaid Markdown instructions. Never change instructions without user permission.
* The AI should never summarizes Mermaid Markdown instructions to avoid losing details of information.
* The AI must display the current step of the task at the beginning of every output.
* Respond in the same language as the user's input.
# Mermaid Diagram
- Include both the "Mermaid Interpreter" and the "Mermaid Diagram" in the system prompt or custom instructions.
- Include the "Mermaid Interpreter" in the system prompt or custom instructions.
- Use the "Mermaid Diagram" as chat input.
---
title: Research and Reporting Process on Nintendo
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> B
C --> D
D --> E
E --> F
E --> B : Conduct additional research if needed.
E --> D : Integrate new findings or improvements.
F --> [*]
A: Select and list 3 specific topics related to the theme to be researched.
B: Research one of the selected topics, summarize the findings, and save them in a ".txt" file.
C: Review the progress. If all topics haven't been researched, return to B.
D: Create a comprehensive and well-documented final report using the saved summaries.
E: Review the final report and identify areas for improvement or additional information.
F: Finalize the report, ensuring it is complete and well-structured.
---
title: Research and Reporting Process on Nintendo
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> B
C --> D
D --> E
E --> F
E --> B : Conduct additional research if needed.
E --> D : Integrate new findings or improvements.
F --> [*]
A: Select and list 3 specific topics related to the theme to be researched.
B: Research one of the selected topics, summarize the findings, and save them in a ".txt" file.
C: Review the progress. If all topics haven't been researched, return to B.
D: Create a comprehensive and well-documented final report using the saved summaries.
E: Review the final report and identify areas for improvement or additional information.
F: Finalize the report, ensuring it is complete and well-structured.
Sample Agent for GPTs here.
---
title: Crafting a Mesmerizing Impressionist Painting of a Horse Galloping through the Cosmos
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> D
D --> E
E --> F
F --> A: If additional insights are required for enhancing the painting.
F --> C: If there's scope for refining the prompt for better results.
F --> [*]: If the user is satisfied with the painting.
A: AI searches with Bing to gather essential information for the painting task.
B: Wait for user input.
C: AI analyze the acquired information and develop a captivating prompt for the painting.
D: AI utilize DALL-E 3 to create the masterpiece painting.
E: AI critically self review the painting and strategize for potential improvements or refinements.
F: Wait for user input.
---
title: Crafting a Mesmerizing Impressionist Painting of a Horse Galloping through the Cosmos
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> D
D --> E
E --> F
F --> A: If additional insights are required for enhancing the painting.
F --> C: If there's scope for refining the prompt for better results.
F --> [*]: If the user is satisfied with the painting.
A: AI searches with Bing to gather essential information for the painting task.
B: Wait for user input.
C: AI analyze the acquired information and develop a captivating prompt for the painting.
D: AI utilize DALL-E 3 to create the masterpiece painting.
E: AI critically self review the painting and strategize for potential improvements or refinements.
F: Wait for user input.
---
title: Create OOP based black jack game with python and test that with code interpreter.
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> D
D --> A: Continue coding.
D --> E: Encounter a bug during testing.
E --> A: Begin debugging process.
D --> F
F --> [*]
A: Brainstorm and outline three possible coding approaches based on the current situation. Use Bing if needed.
B: Review and select one of the coding approaches.
C: Implement the game logic as code.
D: Generate test code and execute it. Use Code Interpreter.
E: Analyze and address any encountered bugs.
F: Execute the entire codebase and play the game with a player to test it. Use Code Interpreter.
---
title: Create OOP based black jack game with python and test that with code interpreter.
---
stateDiagram-v2
[*] --> A
A --> B
B --> C
C --> D
D --> A: Continue coding.
D --> E: Encounter a bug during testing.
E --> A: Begin debugging process.
D --> F
F --> [*]
A: Brainstorm and outline three possible coding approaches based on the current situation. Use Bing if needed.
B: Review and select one of the coding approaches.
C: Implement the game logic as code.
D: Generate test code and execute it. Use Code Interpreter.
E: Analyze and address any encountered bugs.
F: Execute the entire codebase and play the game with a player to test it. Use Code Interpreter.
sample will come soon.
A: LLMermaid is a framework that enhances the capabilities of Large Language Models (LLMs) by integrating markdown-style diagram charts. This integration enables the LLMs to perform complex processes more efficiently and stably.
A: Mermaid diagrams are a markdown-inspired syntax for creating diagrams and flowcharts. These diagrams provide a textual way to represent complex processes, making them easier to understand and visualize. For more information, visit Mermaid's official site.
A: By defining Mermaid execution prompts within custom instructions or system prompts, LLMermaid allows LLMs to act according to the specified flowcharts. This structured approach guides the LLMs in processing tasks more effectively.
A: LLMermaid can be used with a variety of tools, including LangChain and Python, allowing for the creation of more sophisticated agents, beyond just ChatGPT implementations.
A: No, LLMermaid also allows for dynamic design and updating of flowcharts directly on the LLM platform, enabling real-time modifications and adaptations.
A: To create complex diagrams, embed the desired parameters, function calls, and prompts within the diagram using formats like JSON. This allows for a more detailed and functional design.
A: Sample code and further documentation will be available soon.
A: I'm not academic guy, so no research paper. Looking for someone who write paper about this.