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

Tsinghua has built a "game company" with "zero labor content" with ChatGPT, from the boss to the employee is the kind of AI!

As long as you come up with an idea, AI will help you with the complete process from design to testing.

The whole process can be completed in seven minutes at a cost of less than $0.30 (a little more than two yuan)!

The "company" is named "ChatDev", which, as its name implies, is developed through chat.

In the Company, more than a dozen chatbots play different roles, stringing together to complete the development work.

The whole process can be completed with ChatGPT, and there is no need to train the corresponding model for each link.

The average time the "company" takes to develop a game is 409.84 seconds, the fastest is less than three minutes and the slowest is 17 minutes.

In terms of consumption, ChatDev uses ChatGPT's API (3.5-turbo), with an average of about 48.5K token per game.

According to this data, the average cost of developing a game with it is only $0.2967.

So how does the "company" work?

The AI research team was asked to design a set of "ChatChain" connected by ChatGPT for the "company".

Each bot plays different roles such as CEO, programmer and designer, covering all the positions involved in game development.

In order to avoid mistakes or hallucinations, every step in the R & D process is accomplished by two bot.

Specifically, ChatDev needs to go through four major steps: design-programming-testing-documentation when making games.

Before the design process begins, humans will provide a preliminary idea, which is the only place where humans are needed.

The idea will be analyzed and evaluated jointly by CEO, CTO and CP (product) O played by bot.

CEO will discuss with CPO and CTO respectively to determine the presentation of the game (Web / desktop / mobile …) And the programming language used.

In addition, in the design process, the specific role played by each bot is assigned by two instructor (also bot).

In order to improve the quality of design work, the team also introduced two working mechanisms: "memory flow" and "self-reflection".

Memory flow keeps a record of each conversation for bot to read at any time to ensure consistency of ideas.

The "self-reflection" mechanism is to generate a "pseudo-self" when the bot has completed their work but failed to meet the requirements, and feedback the questions and related conversations to the instructor.

After the design work is over, we enter the programming process, including code writing and graphical interface design.

CTO presents requirements and general ideas to programmers, and then programmers write code.

The designer will generate the GUI scheme and call the relevant tools to generate image resources, which will be integrated by the programmer.

The programming process also introduces working mechanisms to improve quality and efficiency, including "code management" and "thinking guidance".

The code management mechanism can save multiple versions of a development project so that it can be rolled back if a problem occurs.

The two points of the thought guidance mechanism are to let CTO and programmers "switch roles" and understand each other's ideas, so that they can better solve problems when they arise.

After the program has been compiled, it is time for testing.

The testing process is divided into two steps: the review of the code and the actual operation, which involves the roles of "code reviewer" and "test engineer".

The testing process also introduces a "thought guidance" mechanism, when the test feedback is not clear, the programmer and the test engineer will switch roles.

After the test is complete, the ontology of the game is done, and the next thing to do is to document.

The document mainly includes two categories: environmental description and user manual.

The former illustrates the environment on which the game depends, which is done by the programmer under the guidance of CTO.

In the latter, the content decided by CEO will be generated by CPO.

At this point, the development of a game is all over.

Apart from providing the initial idea, there is no human shadow in the whole process.

Of course, the dialogue, code and other information in this process are visible to human beings, ensuring the flexibility of development.

If necessary, you can also intervene manually, such as replacing the generated GUI.

The above is a full introduction to the working process of ChatDev.

ChatDev is created under the guidance of Professor Sun Maosong of NLP Lab of Tsinghua University. His research direction is natural language understanding, Chinese information processing and so on.

Associate Professor Liu Zhiyuan of the laboratory is the co-author of the paper. His research interests are knowledge graph and semantic computing, social computing and computational social sciences.

The first author of this paper is Dr. Qian Chen, who graduated from Tsinghua Software Institute. He walked Tsinghua Zhibo from Beijing Institute of Technology in 2016 and worked as an applied researcher at Tencent after graduating in 2021.

One More ThingChatDev uses the role played by AI to simulate the social scenario of the company.

Not only this "AI game company", using AI to simulate human society has become a research trend.

For example, the game version of Westworld, which we have introduced before, uses AI to manipulate NPC to build a society in the game.

And the team used AI to control the characters and made a TV series against the backdrop of South Park.

What kind of environment do you think AI can simulate? Welcome to communicate in the comment area.

Paper address:

Https://arxiv.org/abs/2307.07924

Reference link:

[1] http://nlp.csai.tsinghua.edu.cn/staff/

[2] https://www.linkedin.com/in/qianc62/

[3] https://twitter.com/fablesimulation/status/1681352904152850437?s=20

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.