The programming version of GPT soared to 30 stars, AutoGPT is in danger!

Written by Wang Ruiping

After AutoGPT, GPT family has added GPT-Engineer, a new member.

Like other GPT family members, it has the ability to generate the entire code base, learn your coding style, and be easily adjusted and expanded, all triggered by user instructions. Now, programmers are out of work again.

Easy to use, flexible, and easy to add new AI steps are the best features of GPT-Engineer Strong advantage.

Users can use advanced prompts to let the AI ​​gradually build the user experience and feed feedback back. Over time, the AI ​​is able to remember this feedback.

1. Blast on GitHub: Earn 30k stars

GPT-Engineer debuted on June 11. It was developed by Depict founder and chief technology officer Anton Osika is developed as an AI programming tool for programmers.

In just over a week since its launch, GPT-Engineer has gone viral on GitHub and quickly gained 30k stars.

According to Anton Osika’s tweet, when using GPT-Engineer, you can :

• Generate code base with a prompt word
• Ask clear questions
• Generate technical specifications
• Write all Necessary Code
• Easily add your own reasoning steps, modifications, and experiments
• Lets you complete a coding project in minutes.

2. Advantage: One prompt word generates the entire code base

As mentioned above, the biggest advantage of GPT-Engineer is that it can generate The entire code base.

## Netizens were eager to try it and commented in the interactive area: "This looks too Great, do you want to give it a try?"

The editor can't help but sigh, this brings us one step closer to AGI...

3. Actual test: Completed Snake game project

We also tested the snake game examples listed in the project, and now describe them for everyone to understand:

There are roughly 3 steps to complete the task:

1) Tell GPT-Engineer what you want it to complete;

2) GPT-Engineer asks the user to input ambiguous questions to clarify the task requirements more ;

3) GPT-Engineer starts building and running code.

Input prompt

First, you need to let GPT-Engineer know what to do by entering a prompt.

The prompt word for the Snake game is roughly "a web version of the Snake game that can be played by multiple people." The Python backend adopts a system with MVC components and uses html and js technologies when needed.

Ask questions

Then, GPT-Engineer asks more detailed questions about the task requirements, such as how the snake moves of? How many players can join this game? How often is game status updated?

It is worth noting that GPT-Engineer does not ask these questions unconditionally, but uses a QA approach to identify missing details that need clarification.

Generate game code

After the above issues are clarified, GPT-Engineer can generate the code for multi-player Snake game according to user requirements .

4. Remember the code: store interaction history in a folder

GitHub not only exposes the entire program settings from input to output, but also highlights How the system remembers the code:

Specify the AI ​​ID by editing a file in the folder. (identity)
• Edit the identity and improve it to make the AI ​​agent "remember" the item. (main_prompt)
• The history of communication with GPT-4 is recorded and stored in the logs folder

This special function makes it easier to understand the system Your preferences, improve the efficiency and accuracy of generated code when performing similar operations.

5. Raise questions: Help users fill in missing details

Here, I have to mention the uniqueness of the project, that is, when users enter their needs When asked, GPT-Engineer will not accept it directly, but will ask questions based on its own judgment to help programmers make up for the missing details.

The process is divided into two steps:

(1) Requirements refinement

(2) Software construction

In the requirements refinement phase:

1) The folder containing requirements and problem instructions provided by the user is submitted to GPT-Engineer and placed in the GPT initialization message;

2) The system receives feedback from GPT-4, understands the issues that need clarification and prompts the user;

3) GPT-Engineer loops the process, explaining all issues until GPT -4 until "satisfied".

In the software construction phase:

1) The user requirements extracted in the previous stage are packaged and combined with the GPT system prompts and another Wrap together a set of user output instructions.

2) GPT-Engineer receives the response from GPT-4 and then creates a source code file to provide instructions to the user.

6. Project philosophy: simple and easy to use, providing value to users

GitHub also highlights the main project philosophy of GPT-Engineer:

• Simple and easy to use, providing value to users;

You only need to log in with Google to operate easily, and the entire code base can be generated after entering the prompt word.

• Flexible and easy to add new AI steps;
• Supports advanced prompts and remembers user feedback;

The system can automatically remember your operations, and complete the entire project in your style;

• Quickly switch between human and artificial intelligence;
• All calculations are recoverable and permanently saved in the file system.

7. Summary: AI code generation tools are popular all over the Internet

After AutoGPT, the AI ​​code generation tool GPT-Engineer is popular all over the Internet.

As an AI tool that can generate code according to instructions, it can learn different coding styles and help you complete coding projects in a short time. The whole process consists of the requirements refinement promotion phase and the software construction phase.

The most unique thing about the project is that developers submit requirements in text files. GPT-Engineer does not accept these requirements unconditionally, but asks many detailed questions to allow programmers to clarify deficiencies. details.

8. Comments from netizens: We are also trying an application similar to GPT-Engineer

We collected some comments from netizens under Anton Osika’s post:

“Bonus Features: Overpromise and underdeliver like a true consulting engineer. "

Some netizens also showed similar applications in the comment area: "This is cool, I am developing some similar applications dev-GPT."

The editor also personally tested this: the application can automatically extract like a real Python developer User needs, output the program you want, and gained 153k stars on GitHub.

"GPT-Engineer is a game changer" This quote was written by user OxVivek said. I'm curious to see what innovations it brings compared to smol, and can't wait to complete programming projects in minutes. ”

GPT-Engineer indicates that future creation software will enter a new era of human-computer interaction. In addition, dev-GPT, Auto-GPT, smol, etc. mentioned by netizens can also output your The code you want, you can choose the most suitable application according to your needs.

In the future, you won’t have to rack your brains to write a line of code, and creating a project will be as easy as chatting with a friend.

If you are interested in this and are a programmer, you might as well use Google Register an account and test it yourself, write the program you want in the conversation, create a software system, and leave your feelings in the comment area...

References ：

1.https://github.com/AntonOsika/gpt-engineer

2.https://twitter.com/antonosika/status/1667641038104674306?cxt=HHwWhIC-kYms06QuAAAA

The above is the detailed content of The programming version of GPT soared to 30 stars, AutoGPT is in danger!. For more information, please follow other related articles on the PHP Chinese website!