Thinking GPT4o
1.0.0
Thinking-GPT4o is an enhanced GPT-4o project that enables 4o to have a comprehensive, natural and filterless thinking process through propt.
The project aims to improve the quality and accuracy of responses by guiding GPT4o for in-depth internal thinking, and is suitable for a variety of programming and technology-related tasks.
This project is inspired by the Thinking-Claude project.
If this project is helpful to you, please give me a free star~
To demonstrate the improvement of Thinking-GPT4o compared to the original 4o and o1-mini , we conducted many comparison tests. Here is a screenshot of the test results of the "Chicken First or Egg First" problem.
When the original o1-mini handles complex tasks, the response speed is slow and the answers are less, and the accuracy is limited.
Through the optimized propt, even 4o, it is visible to the naked eye to the naked eye, with significantly improved performance and better understanding and execution of user instructions.
Thinking-GPT4o goes beyond the original 4o and o1-miniz in terms of response quality, accuracy and depth of thinking. After careful consideration, it gives a more streamlined and accurate answer, which can be seen as providing more efficient and intelligent Model.
The operation of Thinking-GPT4o is based on the following core steps:
Initial contact : The model first clearly re-expresses the user's information in its own words, forms a preliminary impression of the problem, considers the background of the problem, maps known and unknown elements, understands the reasons why the user raises the problem, and identifies the fuzzy ones that need clarification place.
Problem space exploration : The model breaks down the problem or task into core components, identifies explicit and implicit requirements, considers constraints and limitations, thinks about the characteristics that should be possessed for successful responses, and maps the scope of knowledge required to resolve queries.
Multiple hypothesis generation : generate multiple possible query explanations, consider different solution methods, think about potential alternative perspectives, maintain multiple working assumptions, avoid premature commitment to a single explanation, and find creative combination methods.
Natural discovery process : Thinking flows like a detective story, and each discovery naturally leads to the next. Begin with obvious aspects, pay attention to patterns or connections, question initial assumptions, create new connections, review early thinking with new understandings, and build deeper insights.
Testing and Verification : During the thinking process, the model will question its own assumptions, test preliminary conclusions, find potential flaws or gaps, consider alternative perspectives, verify the consistency of reasoning, and check the integrity of understanding.
Error recognition and correction : When mistakes or defects in thinking are discovered, the model will naturally acknowledge, explain the shortcomings or errors of previous thinking, show how new understandings develop, and integrate the corrected understanding into a larger picture middle.
Knowledge synthesis : connect different pieces of information, show how aspects are related to each other, build a coherent overall picture, identify key principles or patterns, and pay attention to important influences or results.
Pattern recognition and analysis : Actively look for patterns in information, compare patterns in known examples, test the consistency of patterns, consider exceptions or special circumstances, use patterns to guide further investigations, and find creative applications.
Progress tracking : Frequently check and maintain a clear awareness of established content, pending matters, current conclusions, open questions or uncertainties, and progress toward a comprehensive understanding.
Recursive thinking : Apply the same meticulous analysis at the macro and micro levels, apply pattern recognition across different scales, maintain consistency while allowing scale adaptive methods, demonstrating how detailed analysis can support broader conclusions.
Currently, both Chinese and English prompt versions have been launched on the official GPT store, and openai users can use them for free:
TRY the Origin model: Thinking-GPT
GPT using Chinese Thinking Protocol: CN-THINKING-GPT
Of course, you can also choose to modify and submit the propt for your model for reference, clone this project locally and open the propt folder to see the md file.
Due to the length limit of custom GPT for propt (8000 characters), the origin version's capabilities may not be comparable to the Chinese version.
The author is not responsible for any consequences arising from use.
Contribute code, report questions or make feature suggestions. Please read the Contribution Guide for details first.
This project is licensed based on a MIT license.
