It's hard to keep up with the constant developments in generative AI. It's moving so fast, and that's not down to companies like Microsoft and OpenAI alone. The developer community and enthusiasts all around the world are already starting to stretch what this technology can do. AutoGPT is one such project and is already being hailed as the next big thing.
While claims this is the first indication of AGI (Artificial General Intelligence) may be a little premature and wide of the mark, AutoGPT is certainly a look at what the next likely steps in generative AI are going to be. It's like taking ChatGPT or Bing Chat and letting it run wild. The AI is doing almost all of the work on its own, and that's what makes it interesting. And potentially scary.
AutoGPT has quickly become the hottest repo on GitHub, and anyone can try it out for themselves. It's far from what you'd call a finished product, but it has already spawned a number of interesting projects. The way this stuff is continuing to grow, it's not going to be long before it becomes exceptionally powerful.
What is AutoGPT exactly?
AutoGPT is a project developed initially by an independent developer going by the handle Significant Gravitas. The official description says:
"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 "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."
In simple terms, AutoGPT uses GPT-4 to act autonomously. Where ChatGPT or Bing Chat currently require regular text prompts, AutoGPT can 'think' for itself and generate its own prompts to accomplish its task. Besides this, it can also search the web, has memory, and can write to files, which gives it a significant advantage over other GPT-4 based chatbots.
In an attempt to learn more about AutoGPT I tasked it with teaching me, and it went off, found the following information, and summarized it into the following points.
- AutoGPT is an open-source Python application that uses GPT-4 to perform various tasks, including automated execution of multi-step projects.
- It is an AI agent that can perform web-based actions unattended, assign subtasks to itself, search the web, and improve its own code.
- AutoGPT uses advanced algorithms and natural language processing to generate written content, including blogs and emails.
- It can rewrite and improve its own code, allowing it to recursively debug, develop, and self-improve.
The fact it's able to improve itself without any outside interference is mind-blowing, and perhaps one of the reasons some are comparing it to AGI. Instead of requiring iterations on prompts as you would have to do with ChatGPT, AutoGPT can figure out how to make itself better all on its own.
How to try AutoGPT right now for yourself
AutoGPT is open source and available for anyone to try out for themselves. The official client is built in Python and requires a little bit of work to set up. The GitHub repo walks you through every step, but you'll also need an OpenAI API key as well as a couple of other things.
For the average, casual user, though, this is a pretty intense process. The easier way is to use a tool spawned from AutoGPT called Godmode.
This is AutoGPT but with a more user-friendly design and available through the web browser just the same as ChatGPT. It began totally free but as word got out and demand increased, you'll now either need to register or provide your own OpenAI API key.
If you are using your own key (and you should) be aware that AutoGPT can churn through credit at a vastly increased rate over most other tools, you may have tried. In a morning of playing around with it I used almost $1 of my credit on just two experiments. For more complex projects it also seems like it can keep going and going as it iterates and tries to do better, so you need to keep an eye on it.
But it's really easy to use. As with ChatGPT or Bing Chat, your first step is to say what you want it to do. From there you can add sub tasks that will help AutoGPT figure out how to deliver what you want.
As an extension of my tryout to learn more about AutoGPT, it actually gave me an idea of something I could build myself using Python, an outline of the project and then proceeded to start building and testing the code. Absolutely mindblowing. Compare this to Bing Chat that gave me a summary of what it is and then suggested a few already built projects and you get an idea of how powerful AutoGPT can be. But there comes a point you'll need to reign it in because it can go really deep.
Windows Central's take: Exciting but no replacement yet
Personally, I don't think I'm ready to replace ChatGPT or Bing Chat with this just yet. While its power is impressive, it can be fairly slow since it's iterating each step. I'm also not sure I'm yet ready to just set an AI on a task and leave it to it. Part of my thought process is seeing each response and thinking about how I might make it better or get something else from it.
Nevertheless, it's really exciting and is going to spawn some very interesting projects. If there's a specific task you want to complete that isn't too broad, there's a good chance you'll be able to get AutoGPT to complete it for you.
