AutoGPT very well could be the next big thing in AI. With all of the sensationalist headlines that have been showing up lately, it might be easy to brush this statement off. But after trying this new API out for ourselves, we have to admit that this one is worth looking into.
With the explosion in large language models we have seen recently, artificial intelligence is having its biggest year to date. Every day seems to bring new breakthroughs and advances in technology, and it looks like AGI, or “artificial general intelligence,” is creeping ever nearer.
So, what is AutoGPT? And how does it get us closer to AGI? Let’s talk about it!
What Is AutoGPT: The Next Step in AI Development
Since OpenAI released ChatGPT a few months ago, developers have been tinkering with and tweaking the model to perform some truly incredible tasks. The more recent release of GPT-3.5 and GPT-4 takes the chatbot’s capabilities to another level. But where the real excitement lies is in the API.
You see, if you want to tweak the model and build your own unique projects with it, you interact directly with the ChatGPT API. Doing this lets you craft some really powerful applications. You can tweak and fine-tune GPT with your app to do whatever you want.
©metamorworks/Shutterstock.com
Many devs found unique new ways to create their own chatbots, virtual assistants, web scrapers, and so many other neat things. We’ve stopped trying to keep up with all of the new projects fellow developers were releasing on GitHub because there were just too many — like, hundreds per day.
But the release of AutoGPT really caught the attention of the AI community as a whole. With around sixty thousand stars on GitHub and counting, this open-source project aims to supercharge GPT-4’s capabilities in a way we haven’t seen before.
So what is AutoGPT? Generally speaking, it is one of the first implementations we have seen of an AI Agent.
AutoGPT Is an AI Agent That Can Run Autonomously
An AI agent like AutoGPT has the ability to run autonomously, as the name suggests. This means that, once programmed, the agent can identify a problem, and choose appropriate actions, goals, and states. It can learn how to accomplish those goals until it reaches a logical conclusion.
Things like testing and debugging code, searching for new information on the internet, and proactively planning ahead based on its preprogrammed functions are no longer impossible. AutoGPT accomplishes this by defining a set of up to five goals.
For example, if you wanted to create an agent for autonomous software development, you could easily do this. If you have some spaghetti code that you would like to improve, you would set your agent’s goals to be: read, evaluate, improve, and write tests.
One YouTuber offers a shining example of this capability in action using some simple Python code:
Features of AutoGPT
If software developers did not fear for their jobs before, they are probably starting to sweat a little bit now. But, if you’re like us, you realize the possibilities for creating even more useful programs are endless. And this is due primarily to the advanced feature set of AutoGPT. Let’s talk about that next.
Internet Access for Searches and Information Gathering
When ChatGPT first came out, many users longed for the ability to connect it to the internet. Considering the obvious limitation of GPT’s cutoff date of 2021 for relevant data, this was a valid desire. While a few developers were quick to come up with extensions that provided this functionality, they didn’t always work perfectly.
The good news is that Internet connectivity is one of the strong suits of AutoGPT. It can hop on the web, gather information based on its goals, and report back with only the most pertinent info. In other words, it can autonomously take control of the information-gathering process to spit out results much faster than you could ever type your own query into Google.
Memory Management and GPT-4 Text Generation
Having the ability to recall previous topics and results is invaluable. AutoGPT can store and retain information it has gathered so it can put it to use toward its goals.
While ChatGPT has had this ability to some extent, with the short-term recall of current conversations, the implementation of AutoGPT using GPT-4 means that it can take full advantage of the 8k token limit for even more advanced memory recall.
Access to GPT-3.5 and Other APIs like Google and Pinecone
If you don’t have access to the GPT-4 API yet, you can still use AutoGPT using the older GPT-3.5 model. While not as powerful as GPT-4, the older model is significantly faster and nearly as capable of the same tasks.
You can also connect AutoGPT to the official Google search API for error-free access to the search engine. This does take some extra setup, but the results are worthwhile, especially if you run into issues with search functionality.
Another API that further enhances AutoGPT is the Pinecone API. This allows you to store vector-based memory, enabling you to recall significantly larger file sizes and load only the relevant memories that the agent requires at the time. This requires a Pinecone account, but is certainly worthwhile if you want to tinker with larger file sizes than the standard GPT limit allows.
How to Try AutoGPT for Yourself
If you have played around the OpenAI’s API or other APIs in the past, then this will come as second nature to you. If this is your first foray into setting up an API, then the good news is that the installation and configuration process is still very forgiving.
You can run AutoGPT on the most popular operating systems like Mac, Windows, or Linux. To get started, you will need to have an IDE like VSCode with Dev Container installed. In addition to that, you will need the latest version of Python, which is 3.8 at the time of this writing.
It goes without saying that you will need an OpenAI account and an API key, and any other API keys you want to use, such as the Google or Pinecone API. Once you have the prerequisites, the installation process can be done in a few steps via your Terminal or Command Prompt.
The official GitHub repository offers detailed instructions on how to install and run AutoGPT. You can also follow along with a YouTube tutorial to get up and running if that is more your style.
Implications and Future Possibilities
We’ve already touched on AGI, and how the rapid advancements in language models are pushing us ever closer. AutoGPT looks like the next big leap (at least for this month), and it is showing us things that we never thought were possible.
If you had told a software developer a year ago that they would be able to completely automate parts of their job, they would never have believed you. Now? It looks like whole career fields will be swallowed up by AI.
AutoGPT is a stepping stone toward AGI, and if you’re not paying attention, you might just find yourself surprised when large parts of IT are automated away entirely. But, in the meantime, getting to know this new technology will show you that it opens far more doors than it closes.
Learning to code and build complex applications is a significantly less daunting task with the help of AI, and AutoGPT gives you a huge leg up if you are a developer. Now that we are here, the future of AI looks even brighter. But instead of waiting around to see what happens next, crack open your developer console and try it out for yourself!
The image featured at the top of this post is ©lassedesignen/Shutterstock.com.
