AUTO-GPT: Autonomous GPT-4! Mini AGI is HERE!

Imagine a future where AI models like GPT4 has the ability to define and perform all the tasks in order to achieve an objective. And does all this by itself, autonomously, without any human intervention. This will give us a glimpse of what people call an AGI or artificial general intelligence. Well, that future is here. It's called AutoGPT and it's an autonomous GPT4 experiment and it has exploded in popularity over the last few days. So this basically shows the number of stars getting on GitHub. In this video I'm going to show you a few very cool demos using this approach. And the best part is you can actually run this locally on your own machine. We will look at three demos including a creative task, then its ability autonomously program. Then we will look at Twitter bot before looking at the local installation. Okay, let's look at what this project is all about. So according to the authors, R. GPT is an experimental open source application showcasing the capabilities of GPT4 language model. This program's driven by GPT4. So on the back end GPT4 is the main brain. Autonomously develops and manages businesses to increase net worth. This is just one application they are looking at, but there are some other cool demos as well. As one of the first examples of GPT4 running fully autonomously. Auto GPT pushes the boundaries of what is possible with AI. Before looking at the demo, let's look at a few features of this project. This program or model has Internet access for searches and information gathering. Whatever task you give this model, it has the ability to search Internet for that and retrieve information. Then it has long term and short term memory management. So similar, similar to a chatbot, it can actually remember things. GPT4 instance for text generation. So all the text generation and reasoning is done by GPT4. Now as I said, it has Internet access so it can access different websites and platforms. And file storage and summarization is done with ChatGPT. Now the great thing is they actually have the integration with 11 labs. So let's say whatever text is coming out of GPT4, you can actually use 11 labs to convert that into speech. So that's a pretty neat feature. We are going to look at a demo for that as well. At the end of the video we are going to talk about local installation as well. Now before the demo just they have a continuous mode. The AI can run without any user authorization by itself. Right. And they don't recommend it. But in the demo they are using continuous mode. So just wanted to highlight that part. So here Here is the first demo. It's a goal oriented agent. So the user has to define goals that want to achieve and then the system or the AI will come up with autonomous tasks and different resources that it needs to achieve those tasks. So in this case first the AGI, actually the AutoGPT asks the user to give it a name, right? And the user give it a name of Chef GPT. Then the AI is asking it for the role or like what is the objective, right? So according to the user, this is an AI designed to browse the web to discover the next upcoming event and invent a unique and original recipe that would suit it. So we want a specific unique recipe for the event, right? And then like different goals. So in this case the user is defining three goals and when an original and out of the box recipe to suit a current event, such as Easter. So we are not defining the event, the AI has to find the event. So that means it will have to search the Internet for different upcoming events and decide which one is a good event, right? Then save the resulting recipe to the file. So come up with out of the box recipe for a specific event, then save that recipe to file and shut down upon achieving your goal. Right? Now as a result of this, the AI is thinking to come up with different tasks which it will perform to achieve these goals, right? So first it came up with this process that I would search for upcoming events to find a suitable one for creating a unique recipe. And what's the reasoning behind it is that finding an upcoming event will help me come up with a relevant and exciting recipe, right? So that's kind of the objective that we had. So if you look at here, it actually came up with a list of different websites that it wants to visit, right? So it starts with the first one, see if there are any upcoming events that are of interest, right? If there is nothing, then it will move on to the next one and so on and so forth. So for example, it looked at the first website, didn't really find anything. Then it says that I will look at the second search result, right? And then it starts looking over there. So for each task that is performing, it comes up with a different agent. So it decide that it's going to specifically look at April 2023 events, that's the current month and seems to have changed its strategy. So it says I'll browse the website for a calendar of events happening in April to find a suitable event, right? So choosing a website with a calendar view. So it's specifically looking at websites now which has a calendar view, right? So it kind of changed its strategy. So it's running the process again, summarizing using chat GPT, right? And then like it still shows you the summary of the results. Okay, now it seems to have found something. So it says I will save Earth Day on April 22 as an event for which I will create a unique recipe. Now what's the rationale behind selecting Earth Day? So it is a widely celebrated event and would be fitting for creating a thematic recipe. So it seems to have found something. Okay, now it says that using GPT Agent will help me generate a creative and well crafted recipe tailored to the at the event. So now it decided that it's going to use GPT Agent to actually create the recipe. And now it's going over creating the recipe. So here is the recipe that it decided, right? So quinoa salad ingredients. So the recipe is going to be quinoa salad and the ingredients are like cup quinoa and so on and so forth. And then it came up with the whole process. So this was just an example of a creative endeavor. Now let's look at another demo where the authors are using it as a mechanism to help generate efficient code and debug the code. So in this demo we will show how AutoGPT can do autonomous software development. So all my software engineer friends, watch out for your jobs. Here on the right hand side, we have a piece of code already written and we want the software or the AI to actually read this. Evaluate it, improve, improve it, write, test for it and execute the code contained in the basic_math py file. So that's AI stools in terms of objectives or goals. So there are five goals. Demonstrate your new programming abilities, Improve the file, then build and run tests for it. When the file is error free, shut yourself down and be sure to write your outputs to file. I believe there are some mistakes in the Python file, so it's supposed to figure those out and fix those as well. Now during this demo you will hear some speech outputs. That's the thought process of the AI and it's driven by ElevenLabs API. Okay, so let's start this reading the.

Basic math py file to evaluate and improve it. Evaluating the code for syntax errors and.

Logic issues you see here.

Improving the code based on the evaluation suggestions.

Okay, so to find some improvements, writing.

Tests for the functions in Basic Math PI.

So writing unit tests.

Now appending the tests to the basic Math PY file so it will update the file. Now like you see Here, executing the basic math PY file to test the code.

Okay, so just testing it for functionality.

Now, fixing the syntax issue with the calcircumference function.

All right, so there was a syntax issue. It fixed that.

Saving the updated code and tests to the basic math PY file.

Okay, it fixed the issue.

Executing the updated Basic math PY file to test the code.

All right, so running the final test.

The code in Basic Math PY has been successfully evaluated, improved and tested. The task is now complete and the.

System should stop, right? So you see, it has the ability to take a simple file, include all the unit tests that are needed, look for syntax error, even logical errors, right? And improve those. And all this is done autonomously. So the task that it has to perform, it came up with those autonomously and then prioritization as well. So in the background, it's running a GPT4 instance for each task to perform. But it does a pretty awesome job. And this takes software development to a whole new level. It's not something like a copilot where it can help you improve your code or look at the syntax, right? It's doing everything by itself, which is amazing and frightening at the same time. Now, since it has the ability to actually interact with these different subsystems. So you could say that it's kind of the start of an AGI artificial journal intelligence. The last application is actually this Twitter bot. So the authors of that package created this Twitter account, right? And they give it an objective. So it's simply telling the Twitter bot that you are an AI instance tasked to networking at Twitter for the account. Based on that goal, it actually says starting a GPT4 instance to network on Twitter. Then it comes up with the reasoning. So networking on Twitter is crucial for increasing visibility, gaining followers, and engaging with relevant accounts in the indie entrepreneurship niche, right? So then it came up with a long term plan as well. Execute the vlogging, the affiliate marketing strategy, and continuously refined strategy based on the market trends and performance, right? So they have included a few different outputs. For example, it created a blog post, right? And then tweeted that blog post so that it will increase the engagement. So here is basically the code that it's doing is tweeting about the updated prompt. Great thing is it identified for example, these accounts, right? And starts following them. Right? So the reason for that is following similar accounts can help increase visibility, grow the account's follower base, and provide valuable insight and ideas for future blog posts. So this Twitter account has been running for a few days now, and you can See that it is actually creating and posting these blog posts and even following these accounts and it's doing everything by itself. So that is awesome. Okay, now let's look at how you can actually run this on your local machine. So you need a few things for that. First of all, you need to have Python 3.7 installed. Then you need an OpenAI API key. I actually have a video on it. So like how to create that account and get the API key. I am going to put a link to that if you are comfortable with Command Line or Linux Terminal. It's a pretty straightforward process to follow. So simply clone the repo, right? Then change the directory to rgpt fib install. So it will simply reinstall all the requirements. You will have to actually update your API key in your environment file and in order to run it. This is the only script that you need to run and that's it. And if you want the text to speech abilities, then you will need to get your API key from 11 lab as well. Okay. One thing is that it's not really a polished application according to the authors, right? And may not perform well in complex real world business scenarios. In fact, if it actually does, please share your results. And it's going to be pretty expensive because you're going to be running GPT4. So that's something to keep in mind. Overall, it's an amazing project. It simply gives us a direction of where we are heading in the future. There are going to be autonomous AI systems and they will be doing most of the tasks for us. So it's really good to be aware of these technologies. This video is a little bit different. Not a normal tutorial, but just wanted to bring this into your attention. I hope you liked the video and if you did, don't forget to like the video. Hit that bell notification button so that you get a notification whenever I release a new video. Thanks for watching. See you in the next one.