In just a few days, a new AI has been making headlines and is promising to push the boundaries even further than the groundbreaking ChatGPT.
Developed by Toran Bruce Richards, an American video game developer, this experimental open-source AI called Auto-GPT has been launched on GitHub, a hosting service for software development, and promises to push the limits of what is possible with AI.
A New Disruption in the AI World
On GitHub, we can read:
“Auto-GPT is an experimental open-source application showcasing the capabilities of the GPT-4 language model. This program, driven by GPT-4, chains together LLM [large language model] “thoughts”, to autonomously achieve whatever goal you set. As one of the first examples of GPT-4 running fully autonomously, Auto-GPT pushes the boundaries of what is possible with AI.”
Auto-GPT, based on the GPT-4 model, is capable of autonomously handling a task, even difficult, from A to Z without any human intervention.
Richards responded to an interview with Vice/Motherboard and said:
“Existing AI models, while powerful, often struggle to adapt to tasks that require long-term planning, or are unable to autonomously refine their approaches based on real-time feedback. This inspiration led me to develop Auto-GPT (initially to email me the daily AI news so that I could keep up) which can apply GPT4’s reasoning to broader, more complex problems that require long-term planning and multiple steps.”
In a video he posted, we can see Auto-GPT surfing on Google to find news articles and use them to create a viable business.
Auto-GPT: Autonomous and Independent
Auto-GPT goes further than ChatGPT as it is able to complete a task without any other human intervention than the first prompt, even if this one is vague and not much detailed. If you want to create a website with ChatGPT, you need to give it a lot of precise information including the graphic materials and the creation process involves many back-and-forths between you and the machine. One important thing is that you need to verify what the AI is doing and point out any errors that may occur during the different steps of the process. Auto-GPT removes those steps.
In order to achieve this, Auto-GPT accesses the internet to search and gather useful information and leverages the power of GPT-4 to generate text or code. It can for example go on Google, scan news articles, and select the best data that is needed for its project. It also utilizes GPT-3.5, the previous version of the language model, to store and summarize files. To handle lengthy and difficult tasks, Auto-GPT is thus able to combine multiple language models.
Also, unlike ChatGPT, Auto-GPT is capable of self-evaluating and improving during tasks. For example, it can automatically decide to verify the reliability of websites it visits in order to only use reliable and sourced data. It can also make corrections by itself. It is also aware that its memory is limited and automatically saves its work in a file for future reference which makes it suitable for large-scale projects. Ultimately, Auto-GPT is able to complete projects without the need for programmer intervention other than the initial query.
How Can I Use Auto-GPT?
To initiate a task with Auto-GPT, you must first give it a name, assign a specific role, and explain the five objectives it needs to accomplish. Then, Auto-GPT begins to execute each task one by one until the entire process is completed.
In theory, you don’t need to intervene in the process. There is a continuous mode that enables AutoGPT to autonomously progress in the task without requiring your input. However, during the process, Auto-GPT can seek your permission to proceed to the next step, so that you can monitor the progress of the task.
If you want to give it a try, Auto-GPT’s developer, which goes by the alias ‘Significant Gravitas’ on Twitter encourages other developers to experiment with it. Some said they were able to write a recipe book with it. Traders are also testing it for cryptocurrency purposes.
Latest Demo of Auto-GPT, an open-source experiment to make #GPT4 into a fully autonomous AI🤯🤖
You can now give the AI whatever task you like! 💰
If you would like to give it a try yourself, feel free to check out the GitHub repo in the comments below, it’s free! 👇 pic.twitter.com/sZJx3BQiRS
— Significant Gravitas (@SigGravitas) March 30, 2023
Human Supervision Still Needed
In the same interview with Vice/Motherboard, Richards, however, said that despite the remarkable capabilities AI systems are able to offer, human supervision should remain essential for their responsible and effective use.
“The ability to function with minimal human input is a crucial aspect of Auto-GPT. It transforms a large language model from what is essentially an advanced auto-complete, into an independent agent capable of carrying out actions and learning from its mistakes. However, as we move toward greater autonomy, it is essential to balance the benefits with potential risks. Ensuring that the agent operates within ethical and legal boundaries while respecting privacy and security concerns should be a priority. This is why human supervision is still recommended, as it helps mitigate potential issues and guide the agent towards desired outcomes.”