Shulou(Shulou.com)11/24 Report--

Using your mouth to write code is just around the corner! The AI code generation star project GPT-Engineer became popular overnight, winning 19k stars in just a few days.

After AutoGPT, another star project was born!

GPT-Engineer has become popular all over the network overnight, and the GitHub project has won 19k stars.

This is an AI tool that generates code according to instructions, and you only need to "move your mouth" to build the entire code base directly.

Project address: https://github.com/ AntonOsika / gpt-engineer even, it can learn your code style and allow you to complete the coding project in just a few minutes.

Netizens said that we are one step closer to AGI.

Programmers are going to be eliminated again.

Just a week after a prompt to create a code base, GPT-Engineer went viral on GitHub, attracting many developers to watch.

How charming is it?

Anton Osika, the lead author of the project, first launched GPT-Engineer on June 11 and introduced the biggest features of the AI tool:

-A prompt can generate a code base

-ask questions that need clarification

-generate technical specifications

-write all the necessary code

-easy to add your own reasoning steps, modifications, and experiments

-Open source of the project

-Let you finish the coding every minute.

Imagine that there is no need to write a single line of code in the future, and developers creating a project is as simple as chatting with friends.

This is not just a project, but a glimpse of the future.

GPT-Engineer indicates that software creation will be an era of man-machine conversation in the future.

In addition, according to GitHub, the main project concepts of GPT-Engineer are introduced:

-easy to use and able to provide value to users

-flexible and easy to add new "AI steps"

-support advanced prompts and remember user feedback

-can quickly switch between AI and humans

-all calculations are "recoverable" and persisted to the file system.

What makes this project unique is that the developer submits the requirements in a text file, and GPT-Engineer does not accept these requirements unconditionally, but asks a lot of detailed questions for programmers to clarify the missing details.

The whole process is divided into two stages, namely (1) requirement refinement promotion stage and (2) software construction stage.

The steps of the first phase are:

A text file provided by the user containing the software requirements is submitted to the GPT-Engineer and placed in the initial message of the GPT of the OpenAI, along with instructions to determine and clarify the problem.

The GPT-Engineer system receives feedback from OpenAI GPT-4, understands which requirements need to be clarified, and responds to questions that prompt users to clarify.

GPT-Engineer cycles through this process until all issues are clarified to the satisfaction of OpenAI GPT-4.

The steps of the second stage are:

The requirements extracted in the previous phase are packaged and packaged with OpenAI's GPT instructions (system prompts) and another set of output instructions that GPT-Engineer wants to see (user prompts).

GPT-Engineer received a response from OpenAI GPT-4, and then...

GPT-Engineer creates source code files for software projects that provide instructions for users.

Gluttonous snake has said so much, it is better to try it out. Do a multi-player gluttonous snake web version of the game.

You just need to enter some key tips:

With the Python backend with MVC components, the view needs to pass the state stream to all players and run html and js directly with the code you generate.

For this game code generation, GPT-Engineer will let you continue to answer a few detailed questions.

Rules and mechanisms of the game: exactly how Snakes move, grow, and interact with other players, are there any power-ups, or special game elements?

Player Interconnection: how many players can join the game, is there a game hall or a match?

Game status update: how often do you update the game status, how often do you transfer the status stream to the player, and what is the delay rate?

User interface: what does the game interface look like? Do you have any special design needs and preferences?

Game control: how do players control snakes? Are there any other controls or settings?

Game end status: how does the game end? Is there a scoring system or ranking?

Code structure: are there any specific requirements for the organization of Python backend, HTML, and JavaScript code?

When you have answered these questions briefly, based on this information, GPT-Engineer gives a list of files, classes, and methods needed to achieve your goals.

The concrete steps came quickly.

The last perfect gluttonous snake game code is completed.

Netizens said that "clarifying the problem" is what really makes GPT-Engineer stand out, because fixing problems in the generated code often takes more time than writing the code itself. This seems to be a big step in the right direction.

I wish I had such a weapon when I was at school.

Some netizens are confused about the difference between GPT-Engineer and GPT-4.

However, despite the popularity of the project, a netizen tested GPT-Engineer and gave the experience:

Difficult to set up

It doesn't actually create a file, it just allows you to run it once from the command line (that is, it can't be debugged)

Equal to the ability of ChatGPT4.

Finally, it is concluded that the ideal is very plump and the reality is very bony.

Some netizens also pointed out problems with the benchmark.

Gpt-engineer does a few simple programming tasks and uses them as a benchmark. It can quickly develop a currency_converter for you, but it doesn't code a pomodoro_timer correctly (not yet! )

The installation method is either stable version (stable) or development version (development).

For stable versions:

-pip install gpt-engineer

For the development version:

-git clone git@github.com:AntonOsika/gpt-engineer.git

-cd gpt-engineer

-make install

-source venv/bin/activate

Set up

Run with the API key with GPT4 access:

-export OPENAI_API_KEY= [your api key]

Running

Create an empty folder. If you are in repo, you can run:

-cp-r projects/example/ projects/my-new-project

-fill in the main_prompt file in the new folder

-run: gpt-engineer projects / my-new-project

Result

-check the files generated in projects / my-new-project / workspace

Limit

Using additional thought chain tips, such as Reaffon, should make it more reliable without missing out on the functionality requested in the main tip.

You can specify the "identity" of the AI agent by editing the files in the identity folder.

Editing identities and improving main_prompt are ways for agents to remember things between projects.

Each step in steps.py stores its conversation history with GPT4 in the logs folder and can be rerun using scripts / rerun_edited_message_logs.py.

Anton Osika, lead author of the project.

Anton Osika is the chief technology officer of Depict.ai. This is a start-up that creates machine learning recommendation systems to understand products through cutting-edge machine learning, computer vision and natural language processing.

Just four hours ago, he posted a message saying that GPT-Engineer had jumped to 17k, showing no sign of slowing down.

Reference:

Https://github.com/AntonOsika/gpt-engineer

Https://twitter.com/rowancheung/status/1670829363137499137

Https://twitter.com/antonosika/status/1667641038104674306

Https://twitter.com/antonosika

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.