This article first published the WeChat public account: The fully armed elder brother
The difference between AutoGPT and ChatGPT
How AutoGPT works
Example of using AutoGPT
Installation and use of AutoGPT
AutoGPT obvious ok
When we were still thinking about GPT4, it came!
AutoGPT
is currently an open source, powerful tool that greatly enhances the capabilities of GPT4.
If it is said that when using ChatGPT in the past, it was necessary to continue to use prompts to complete the goal step by step, then AutoGPT only needs to put forward the demand, and can perform tasks with almost no human intervention.
For example, we use VS Code+ChatGPT to write code, and we need to constantly select a certain piece of code to communicate with ChatGPT, and then manually debug. Now we only need to tell AutoGPT to write a certain function, it can automatically write code, debug code, and find solutions to bugs from the web.
The main difference between AutoGPT and ChatGPT
Both AutoGPT and ChatGPT are built on the same technology, and AutoGPT uses the API of GPT3.5 and GPT4 in combination, but AutoGPT is more powerful. The biggest difference between them is that AutoGPT does not need prompts to complete the task, while ChatGPT needs step-by-step prompts to complete the goal.
For example, if we want to use ChatGPT to plan our own trip, we need to give a prompt, such as "Help me plan a trip to Zhangjiajie". ChatGPT then generates a list of things we should consider, including how to travel, where to stay, where to eat, what to see. However, we need to prompt ChatGPT for each subsequent step, such as travel mode or diet.
In contrast, AutoGPT can self-prompt and solve each problem. For example, if we ask AutoGPT to plan a trip, it can search online, and then choose the travel method, travel route, accommodation, and even order high-speed rail tickets for you! (if you grant it permission)
How AutoGPT works
AutoGPT adds the function of AI agent based on ChatGPT. By granting access to the AI agent, it is able to perform tasks in accordance with the given permissions.
For example, an AI agent cannot purchase items on your behalf if it is only authorized to access the internet to search for information. If the AI agent is granted installation permissions, it can help install software.
In order to be able to perform according to user expectations, the AI agent will prompt for permission after each step. Only with permission can the AI agent complete the corresponding action.
AutoGPT actually adds a layer of intelligent agent to GPT. The agent receives user instructions and tells GPT what to do in order to complete the task.
In order to complete the task, the intelligent agent continuously prompts the GPT and receives feedback from the GPT, and then proceeds to the next step of planning. Through continuous iteration until the task is completed. It's a bit like a nesting doll, calling itself by itself.
Example of using AutoGPT
- AutoGPT finds that the user doesn't have Node, Googles how to install it, downloads it, extracts it, and spawns the server, while the user just watches AutoGPT work
2. Auto-GPT is able to write and execute its code using GPT-4, allowing it to recursively debug, develop and improve itself.
For the video, see the WeChat public account: Fully Armed Senior Brother
How to install AutoGPT?
AutoGPT needs to call the OpenAI interface, so, first of all, you need an OpenAI account. Newly opened accounts have a free token of $18, which is enough to experience AutoGPT.
Using AutoGPT requires:
Python3.8 and above (https://www.python.org/downloads/)
OpenAI API (https://platform.openai.com/account/api-keys)
GPT4 API (https://openai.com/waitlist/gpt-4-api)
PINECONE API (https://www.pinecone.io/)
ElevenLabs API(https://beta.elevenlabs.io/)
If you want AutoGPT to speak, you need ElevenLabs API.
Installation steps [2]:
In the CMD, Bash or Powershell window, clone the repository (you can also download the zip package directly):
python git clone https://github.com/Torantulino/Auto-GPT.git
Go to the project directory:
python cd 'Auto-GPT'
Install the required dependencies:
pip install -r requirements.txt
Configure AutoGPT:
Find the .env.template file in the main folder, copy this file and remove the extension of .template cp .env.template .env
Open the .env file in a text editor, locate this line OPENAI_API_KEY= and enter your own OpenAI API,
here you should pay attention, there can be no spaces or double quotes before and after the = sign. If you need other services, add other API Keys and save the .env file
How to use AutoGPT?
Run the script:
On Linux of Mac: ./run.sh start
On Windows: ./run.bat start
To perform an operation:
Authorize a single command, enter y
to authorize N consecutive commands, enter y -N
to exit the program, n
Next story AutoGPT
AutoGPT calls its own mode, and it feels like there will be a problem like recursion, that is, it will fall into a loop. When AutoGPT cannot find a solution to the problem by searching the network, it will keep searching, resulting in an endless loop.
Reference content:
[1]https://autogpt.net/auto-gpt-vs-chatgpt-how-do-they-differ-and-everything-you-need-to-know/
[2] https://github.com/Significant-Gravitas/Auto-GPT