The development of Artificial Intelligent technology has been increasingly massive lately. The enthusiasm of people to create artificial intelligence that can outperform the humans in any way is very large. But until now, various AI technologies are still at the Narrow AI stage where an AI model has the task of solving specific problems such as ChatGPT, DALL E, Deep Fake, etc. It’s no longer a secret that everyone is competing to create an Artificial General Intelligent (AGI) agent which is namely an AI agent that represents the cognitive skills of a human. An AGI is expected to be able to connect, assess, and consider an event just like a human being. It is this potential for AI that Elon Musk worried about many times in his interviews where he feels that Artificial Intelligence is far more dangerous than nuclear. In addition, there are no clear regulations regarding the use and development of AI in society. It makes perfect sense, even as a regulator, that you must be confused about determining the standard limits and freedom of application of AI technology in society, considering that the development of AI is currently still in the experimental stage.
WHAT IS AUTO-GPT
In short, Auto-GPT is an early AGI wanna-be. The Auto-GPT mechanism mimics how AGI works, but is limited on the text output. Auto-GPT is an open-source project created by Toran Bruce Richards, a game developer, that uses OpenAI’s text-generating models which are currently the GPT-3.5 and GPT-4 models to perform the tasks by imitating the cognitive functions of a human including thinking, reasoning, and giving proposed action plan. The main idea in the development of Auto-GPT is to create an AI agent that can respond to input from the user (or master) by involving critical thinking and planning solutions and then elaborating on the inputted topic automatically with the provision of memory management. The information that is entered in the memory can be recalled as material for consideration to determine the next action decision.
HOW AUTO-GPT WORKS AND ITS FEATURES
Instead of using ChatGPT by providing one task or prompt at a time, Auto-GPT performs cognition functions automatically and continuously to complete the tasks that the user provides. Auto-GPT performs the cognition function continuously until the user terminates the Auto-GPT agent.
Figure . Auto-GPT asking for user input
There are two modes to run the Auto-GPT, by default, which are the manual mode and continuous mode. In manual mode, each generated action plan must receive a user authorization to be executed or not, or by adding feedback. As for the continuous mode, Auto-GPT executes any action without the user consent.
Auto-GPT can access directories, read or write a file, and can even execute a source code file in the Python language that it produces.
Figure . Auto-GPT asks for user authorization to write to a file
Auto-GPT is equipped with the ability to browse the internet using the Google Chrome binary to be able to get the actual sources of information. Meanwhile, the text data that is obtained from ChatGPT is limited to a certain period. Auto-GPT supports Text-to-Speech using the API of Eleven Labs which enables the Auto-GPT to speak as if it is your personal AI assistant. Auto-GPT also supports Image Generation.
CONCLUSION
Even though the Auto-GPT project is currently still experimental, we believe that this is a clear evidence of the early rise of Artificial General Intelligence (AGI) because all industries are competing towards the development of Artificial Intelligence. The installation of using the Auto-GPT is very easy; we described it in detail in the following articles:
- How to setup Auto-GPT on Windows
- How to setup Auto-GPT on Linux
- How to use Auto-GPT with Discord
