Machine Heart report
Editors: Du Wei, Chen Ping
Andrej Karpathy of OpenAI touts AutoGPT as the next frontier in prompt engineering.
Recently, there seems to be a new trend in the AI world: autonomous artificial intelligence.
This is not out of nowhere, and recently a study called AutoGPT began to enter the public eye. Andrej Karpathy, Tesla's former AI director and just returning to OpenAI, also promoted it, praising it on Twitter: "AutoGPT is the next frontier of prompt engineering."
Not only that, but there are also claims that ChatGPT is obsolete and AutoGPT is the new player in the field.
Once the project was launched, it won 27K+ stars in just a few days, which also verified the popularity of the project.
GitHub address: https://github.com/torantulino/auto-gpt
The question is, what exactly is AutoGPT? It is an experimental open-source application that demonstrates the capabilities of the GPT-4 language model. The program is powered by GPT-4 and can autonomously achieve any goal set by the user.
Specifically, AutoGPT is equivalent to giving GPT-based models a memory and a body. With it, you can give a task to an AI agent, let it autonomously come up with a plan, and then execute the plan. It also features Internet access, long- and short-term memory management, GPT-4 instances for text generation, and file storage and summary generation using GPT-3.5. AutoGPT can be used to analyze the market and propose trading strategies, provide customer service, conduct marketing, and other tasks that require constant updating.
As netizens say, AutoGPT is taking the internet by storm, and it's everywhere. Soon, there were already netizens who started experimenting, and the user asked AutoGPT to build a website, and AutoGPT was successful in less than 3 minutes. During this period, AutoGPT used React and Tailwind CSS, all on its own, without human intervention. It seems that programmers really don't need to code anymore.
The user later added that his goal was simple: to create a website with React. The request was: create a form, add the title "Made with autogpt", and then change the background to blue. AutoGPT successfully built the website. The user also said that AutoGPT would perform better if it gave more prompts.
Source: https://twitter.com/SullyOmarr/status/1644160222733406214
Let's look at another example. Pretend to run a shoe company yourself, give AutoGPT the order to conduct market research on waterproof shoes, then ask it to give a top5 company and report the advantages and disadvantages of competitors:
First, AutoGPT went directly to Google and then looked for companies with a comprehensive evaluation of waterproof shoes. Once it finds the relevant link, AutoGPT asks itself questions such as "what are the advantages and disadvantages of each shoe, what are the advantages and disadvantages of each top 5 waterproof shoes, the top 5 waterproof shoes for men", etc.
After that, AutoGPT continues to analyze various other websites, combined with Google searches, to update the query until it is satisfied with the results. During this period, AutoGPT is able to determine which comments are likely to be biased towards forgery, so it must verify the commenter.
In the process of execution, AutoGPT even derives its own child agents to perform the task of analyzing the website and finding solutions to problems, all on its own.
As a result, AutoGPT gave a very detailed report of the top 5 waterproof shoe companies with the advantages and disadvantages of each company, in addition to giving a concise conclusion. The journey took only 8 minutes and cost 10 cents. There was also no optimization at all during the period.
How does this AutoGPT, which can complete tasks independently, work? Let's move on.
AutoGPT: Build your own AI assistant in 30 minutes
As an AI agent that has swept the Internet, AutoGPT can be set up in less than 30 minutes. You can have your own AI to help complete tasks and be more productive.
This powerful AI tool autonomously performs a variety of tasks, and the ease of setup and start-up is a key feature. Before you begin, you'll need to set up Git, install Python, download the Docker desktop, and get an OpenAI API key.
Clone the repository
Start by cloning the AutoGPT repository from GitHub
Use the following command to navigate to the new folder Auto-GPT.
Configure the environment
In the Auto-GPT folder, locate the .env.template file and insert the OpenAI API key. Then copy the file and rename it to .env.
Install the Python package
Run the following command to install the required Python packages.
Run Docker
Run Docker Desktop, you don't need to download any containers, just keep your program active.
Run AutoGPT
Run AutoGPT by executing the following command.
Set goals
AutoGPT is a powerful tool, but it's not perfect. To avoid problems, it's a good idea to start with a simple target, test the output, and adjust the target to your needs, such as ResearchGPT above.
However, if you want to unlock the full potential of AutoGPT, you'll need GPT-4 API access. GPT-3.5 may not provide the required depth for the agent or response.
AgentGPT: Deploy autonomous AI agents directly in the browser
Recently, another developer has launched a new exploration attempt on AutoGPT, creating a project that can assemble, configure and deploy autonomous AI agents in the browser - AgentGPT. One of the project's main contributors is Amazon software engineer Asim Shrestha, who has earned 2.2k Stars on GitHub.
- Project Home:https://agentgpt.reworkd.ai/
- GitHub address: https://github.com/reworkd/AgentGPT
AgentGPT allows you to name your custom AI and make it do whatever you want. Custom AI thinks about the task to be done, performs the task, and learns from the results in an attempt to achieve a goal. Here's an example demo: HustleGPT, which sets the goal of starting a startup with only $100 in funding.
Another example is PaperclipGPT, which aims to make as many paper clips as possible.
However, users will also need to enter their own OpenAI API key when using the tool. AgentGPT is currently in beta and is working on long-term memory, web browsing, and interaction between websites and users.
How much room GPT has for imagination remains to be seen.
Reference link: https://medium.com/@tsaveratto/autogpt-your-personal-ai-assistant-in-less-than-30-minutes-f864fe1599fe