AutoGPT: This Is ChatGPT Supercharged!

Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to see how ChatGPT just got supercharged. Normally, this is OpenAI's chat assistant, where we get a little text box, enter our questions as a text prompt, and it shall answer. However, get this. Today we don't even need to prompt ChatGPT anymore because. And now hold on to your papers, because it can prompt itself. What? How? That sounds insane. But yes, all that is true. We can describe a complex task and it will by itself break it down into smaller steps and it does them all. For instance, here it is asked to buy a pair of AirPods. And as you see, it knows that it first has to search for it look for deals. Yes, we got one. It is indeed on the right track and then gets distracted by a privacy policy. Or perhaps it knows something we don't know yet. Yes, this is a version of ChatGPT that can make decisions and take action. So what can we use this for? Well, we can even ask it to build a website for us and it adds the required goodies step by step. Really cool. Now this is Two Minute Papers, so we want to give it a scholarly task. Oh yes, let's ask it to research itself. That is one more task with several steps involved. And yes, this interface gives us a little opportunity to pop the hood and hear its thoughts. It first wants to search for AutoGPT, browse around a little, and finally create an outline about itself for us. Once again. The key part is that it can take a complex task and break it down into small steps. That is incredible. What I also loved is that it also reminds itself not to wander around too much if it finds a tasty privacy policy and focus on the main task instead. Good thinking, Little AI. Now let's see what happens. Then the search takes place, we get a ton of results, and finally it enters GitHub, the home of AutoGPT. We see how it reads through the page and finally we get our answer. Now, my first question was, is this a copy paste from the website or did it really read the whole thing and and distilled down the information into one paragraph? And yes, it did. Good job, little AI. So at this point my question was, what is AutoGPT? What role could this play in our lives? And have a look at this. We can also get a progress report where it tells us what it has done and what is left to be done. So if yes, AutoGPT feels very much like a team of assistants. The first version of OpenAI's ChatGPT is barely a few months old and it can already prompt itself, chain prompts together and report back to us on where exactly it is. That is incredible. And we can even ask it to have a look at a piece of computer code. Then it again breaks down the task into small evaluation, improvement, and finally testing. And it does all of that like a team of assistants would. Also, this is a computer that kind of fixes itself. How cool is that? What a time to be alive. Now a few more things that you should know about this. 1. By default it asks for permission before each step. That is a good idea. 2. These are just the first few steps and when you start using it you will find that it is a bit slow, a bit expensive and quite limited. Also, I would like to make sure not to overstate what AutoGPT can do at the moment. However, can you even imagine what this will be able to do? Two more papers down the line. If you have some ideas, let me know in the comments below. So do you wish to try it? And can you? Yes, you can. You can run it locally or even try it on the web. I've included both links in the video description. Note that you will need an API key to perform that. Now look. What is this? Who are these people? Are you thinking what I am thinking? Yes, that's right. This is a long line of you fellow scholars and entering an incredible event in London with our longtime partner Weights and Biases. And I also flew there to hold the first two minute papers meetup of sorts ever. Here you see the footage I took where you fellow scholars expressed what you really want. Yes, I approve. And here are some fellow scholars who were very happy to get what they wanted. What a time to be alive. The event was incredible. The energy was electric. I handed out hundreds of papers and gifts and got back what feels like a hundred hugs. Some crushed my ribs, but that's okay. And what you see here is just a portion of the thousand plus people who came. It was so amazing to see your faces and say hello and hear your stories. I really hope to have the honor of doing this again someday. So I would like to send a huge thank you to Weights and Biases for organizing this. I was treated like a king there. And just one more incredibly important thing. Please don't forget you make the papers happen. This show could not exist without you. So thank you so much for everything. You gave me the opportunity to do what I absolutely love and every single day of my life and I could not be happier. Thank you.

Weights and Biases provides tools to track your experiments in your deep learning projects. Using their system, you can create beautiful reports like this one to explain your findings to your colleagues better. It is used by many prestigious labs including OpenAI, Toyota Research, GitHub and more. And the best part is that Weights and Biases is free for all individuals and academics and open source projects. Make sure to Visit them through wnb.com papers or just click the link in the video description and you can get a free demo today. Our thanks to Weights and Biases for their long standing support and for helping us make better videos for you.

Thanks for watching and for your generous support and I'll see you next time.