After AutoGPT, another star project was born!
GPT-Engineer has become an overnight sensation all over the internet, and the GitHub project has already gained 19k stars. It is an AI tool that generates code based on instructions, and you only need to “move your mouth” to build the entire codebase directly. It can even learn your coding style in just a few minutes, allowing you to handle the coding project.
Netizens say that we are one step closer to AGI. GPT-Engineer has been wildly popular on GitHub in just one week, attracting many developers to watch.
How attractive is it? The primary author of the project, Anton Osika, debuted GPT-Engineer on June 11 and introduced its best features:
Generates a codebase with a single prompt
Asks questions that require clarification
Generates technical specifications
Writes all the necessary code
Easy to add your reasoning steps, modifications, and experiments
Get your coding done in minutes
Imagine a future where you don’t have to write a single line of code, where creating a project is as easy as chatting with a friend. It’s not just a project; it’s a glimpse into the future. GPT-Engineer heralds an end in which software creation will be a human-computer conversation.
In addition, according to GitHub, the main project concepts of GPT-Engineer are:
Ease of use and the ability to provide value to users
Flexibility and ease of adding new ‘AI steps.’
Support for advanced prompts that can remember user feedback
The ability to quickly switch between AI and humans on the fly
All calculations are ‘recoverable’ and persistently saved to the file system
This project is unique because developers submit requirements
In a text file. Instead of accepting them unconditionally, GPT-Engineer asks many detailed questions to get programmers to clarify missing details. The whole process is executed in two phases, namely (1) the requirements refinement facilitation phase and (2) the software build phase.
The steps of the first phase are:
A user-supplied text file containing the software requirements is submitted to GPT-Engineer and placed in the initial message of OpenAI’s GPT, along with instructions for identifying clarifying questions.
The GPT-Engineer system receives feedback from the OpenAI GPT-4 about which requirements need clarification and responds to the questions, prompting the user for clarification.
The GPT-Engineer loops this process until all issues are clarified to satisfy the OpenAI GPT-4.
The steps in the second phase are:
The requirements defined in the previous phase are packaged and combined with OpenAI’s GPT instructions (i.e., system prompts) and another set of output instructions (i.e., user prompts) that the GPT engineer wants to see.
The GPT-Engineer receives a response from the OpenAI GPT-4, and…
GPT-Engineer creates the source code files for the software project for which the user provides instructions.
To install, choose either the stable version or the development version. For the regular version, run the following:
- Pip install gpt-engineer
For the development version, run the following:
- git clone firstname.lastname@example.org:AntonOsika/gpt-engineer.git
- cd gpt-engineer
- make install
- source venv/bin/activate
To run with the API key with GPT4 access, run the following:
- export OPENAI_API_KEY=[your API key]
Create an empty folder. If it is in the repository, run the following:
- Cp -r projects/example/ projects/my-new-project
- Fill the new folder with the main_prompt file
- Run: gpt-engineer projects/my-new-project
Check the files generated in projects/my-new-project/workspace.
Additional thought chain prompts, such as Reaffon, can make the results more reliable and ensure that features requested in the primary prompt are noticed.
You can specify the “identity” of an AI intelligence by editing the file in the identity folder.