news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Machine Heart Report

Synced Editorial Department

Of course, the next generation of AutoGPT is still open source.

Remember the "AutoGPT" project that AI expert Andrej Karpathy promoted last year? It is an experimental open source application driven by GPT-4 that can autonomously achieve any goal set by the user, showing the development trend of autonomous AI.

In more than a year, the total number of GitHub stars for the project has exceeded 160,000, which shows its continued popularity.

GitHub address: https://github.com/Significant-Gravitas/AutoGPT

Yesterday, the project author announced the arrival of the next generation of "AutoGPT" on social media X, which is currently in Pre-alpha version. Compared with the previous generation, the next generation of "AutoGPT" makes it easier than ever to build, run and share AI agents, while greatly improving reliability.

Image source: https://x.com/SigGravitas/status/1812803289739633018

The author shows how to use the next generation of "AutoGPT" to quickly build, deploy and use Reddit marketing agents that can automatically reply to comments and perform other complex tasks. The new system is no longer as unreliable and inefficient as the first version of "AutoGPT".

When asked whether the project could build multi-agent driven applications, the author answered in the affirmative and stated that this was an important reason for building the project.

Now you can design a graph where multiple expert agents can collaborate to solve a problem. SubGraphs are coming soon to help make it easier to work with complex graphs.

The project also allows for custom addition of nodes, and it is very simple to manually write new blocks (a key feature mentioned below). The author found that Claude did an excellent job in this regard and can create custom blocks perfectly.

Of course, the two main components of the project (the Server and Builder mentioned below) can run on different machines.

However, for the next generation of "AutoGPT", the author admits that it is still in a very early stage, has defects and is relatively basic, but still hopes to share it with everyone and open source it from the beginning.

Main components and key features

The project has two main components, namely the AutoGPT Server at the back end and the AutoGPT Builder at the front end. The Server is responsible for creating composite multi-agent systems, using AutoGPT agents and other non-agent components as their primitives.

• AutoGPT Server (Backend):
• https://github.com/Significant-Gravitas/AutoGPT/tree/master/rnd/autogpt_server
• AutoGPT Builder (Frontend):
• https://github.com/Significant-Gravitas/AutoGPT/tree/master/rnd/autogpt_builder

The specific steps to set up and run the Server and Builder are as follows:

• Navigate to the AutoGPT GitHub repository;
• Click the Code button and select Download ZIP.
• Once downloaded, extract the ZIP file to a folder of your choice;
• Open the unzipped folder and navigate to the "rnd" directory;
• Enter the "AutoGPT Server" folder;
• Open a terminal window in the "rnd" directory;
• Find and open the README file in the AutoGPT Server folder;
• Copy and paste each command in the README into your terminal (Important: wait for each command to complete before running the next one);
• If all commands run without errors, enter the last command "poetry run app";
• In the terminal, you can see that the server is running;
• Navigate back to the "rnd" folder;
• Open the "AutoGPT builder" folder;
• Open the README file in this folder;
• Run the following command in Terminal:

npm install

npm run dev

Once the front-end is running, click the link to navigate to localhost:3000.

• Once the frontend is running, click the link to navigate to "localhost:3000";
• Click on the Build option.
• Add a few blocks to test functionality;
• Connect blocks together;
• Click Run.
• Check your terminal window. You should now see that the server received the request, is processing it, and has executed it.

By following the steps above, you can successfully set up and test AutoGPT.

Video source: https://github.com/Significant-Gravitas/AutoGPT/tree/master/rnd/

In addition to the two main components, the key feature of the next generation "AutoGPT" is the use of "Blocks" to build intelligent agents. You can combine some highly modular functions to create custom behaviors.

Currently, the project has provided corresponding blocks for operations such as Reddit posting, Discord message sending, and Wikipedia summary acquisition. At the same time, the design strives to be easy to create and use. The following is an example of a block for Wikipedia summary acquisition:

class GetWikipediaSummary(Block):

class Input(BlockSchema):

topic: str

class Output(BlockSchema):

summary: str

def **init**(self):

super().__init__(

id="h5e7f8g9-1b2c-3d4e-5f6g-7h8i9j0k1l2m",

input_schema=GetWikipediaSummary.Input,

output_schema=GetWikipediaSummary.Output,

test_input={"topic": "Artificial Intelligence"},

test_output={"summary": "Artificial intelligence (AI) is intelligence demonstrated by machines..."},

def run(self, input_data: Input) -> BlockOutput:

response = requests.get(f"https://en.wikipedia.org/api/rest_v1/page/summary/{input_data.topic}")

summary_data = response.json()

yield "summary", summary_data['extract']

The author said that this is just the beginning, and more blocks will be added in the future, and the UI will be improved to greatly enhance the overall experience and functionality.