What can you do with Autonomous Agents? Auto GPT & Replit = The fastest way to spin up an Agent

[Music] foreign [Music] autonomous agents well they're pieces of software that use llms to chain together thought so they can come up with a question and try and solve it and in fact this new paradigm of AI software is a real real Boon for those of us that do any kind of research or want anything done because these autonomous agents have access to your file system and have access to your local machine now that might terrify a few of you and certainly terrified me for just a fraction of a second before I remembered that each Rebel you spin up is its own computer in the cloud so if you're using an autonomous agent on replit you really haven't got much to worry about because the only thing it can damage is itself in fact the big boy that was released last week is auto GPT Auto gpk comes in a GitHub repository and honestly it's a bit of a pain to set up if you were doing it machine well thanks to our developer Advocate we have a fantastic new template where you can spin up Auto GPT in minutes let's give it a go and explore what an autonomous agent can actually do for us now of course the first thing we have to do is go to openai login and get our openai key you can do that by going to your profile clicking on view API keys and hopefully if you've created one you know what it is otherwise click create new key give it a name click create secret key and copy this now don't forget that if you sign up for a free account your first six months you get a number of dollars of credit to use for your API calls for free otherwise you will have to connect this to a credit card back on replica and let's see how easy it will take GPT template when we get there we're simply going to Fork it make sure to make it private if you want your open AI key to stay private and the forking process happens reasonably quickly now we do have a readme.md file to tell you what to do to set it up but actually the main thing we've got to do is go to secrets and put in our openai API key with that we can click run and we've actually got Auto GPT working already you see it's asking us for us to continue the previous model which was an example one I'm just going to say no to that and you'll see the first thing it prompts me is to give it a name so I know what I'm looking for well let's have a think about what these autonomous agents are good for they're very good for doing independent research and just getting on with stuff they're very good at finding and creating stuff and coming to a conclusion from using Google and using a web browser like a human would but it's best at chaining things together and creating questions to generate an answer for you let's start with something quite simple finding the best headphones for a certain budget so give it its main goal and in this case my main goal is going to be to find the best wireless headphones for under 200 and then it's going to ask me for a couple of goals so first goal would be to find the best headphones all round for under two hundred dollars that are Wireless goal two maximize battery life should have noise cancellation be comfy summoned with a big head should be aesthetically pleasing now autonomous air agents do spend a lot of time thinking and this is because they're coming up with questions to meet the prompt and meet the needs of your topic you'll see here what it does is it comes up with a question what it wants to do next some reasoning and some of its own criticism all you need to do is say yes or no to whether it should follow that prompt so I'm going to say yes and let it do its business you can see that it goes off and gets articles goes off and searches the web and then does more thinking and more analysis based on that now you might be wondering why it's asking me each time to authorize its request and the reason for that is as you can imagine on autonomous AI agent sending requests back and forth to open ai's API gets quite expensive quite fast this model is running on GPT 3.5 so it's reasonably inexpensive but you do get the best results from running out of gpt4 gpd4 is a very expensive way of running this and if we just said yet go for it let it run for as long as it wanted we could rack up a bill without really realizing I'm going to try now to tell it to carry on and just keep going for a few I'm gonna give it 10 more requests so you'll see that using Y and then a number flag allows me to tell it how many times just get on with it and I just said go through 10 all iterations go through 10 more thought processes and see where you get now autonomous agents are not perfect they are a brand new technology in fact they're really a brand new paradigm in AI software that means at the moment they do need a little bit of massaging a little bit of help and they can go down rabbit holes when I first encountered these I asked it to do a big research project on which was the best computer science GCSE in the UK and it went that a massive Rabbit Hole of degree level qualifications before I stopped it and prompted it to get back on the right track these things need to be babied a lot at the moment but they are only a week old so I think that's fair okay so it's shutdown has it come to a conclusion so let's see what it finally decided after buying a distant research on each option including battery life and comfort someone with a big head determined that the Audio Technica ath mx50x bt2 is the best choice not only does it offer 40 hours of battery life but also known for its level of comfort even for people with a big head that's me additionally reviews are saying the overall sound quality is excellent now that's great I mean when I go on a shopping spree and I want the best of something I normally spend hours and hours of my life researching things under certain criteria comparing reviews this autonomous agent might not be the most efficient thing in the world but it's gone off and done something like that and found me a result on its own whilst I could be doing something else and something more productive with my time this is an amazing new technology but let's see what else it can do okay I've got something going on here that I've always wanted to do but never really had the time to achieve I've created a folder in my wrapper called audio and I filled it full of a bunch of MP3 files now something I'd really like to have in a rebel is an RSS podcast feed that I can bring into my phone of audio files that I just drop in myself I've never really had the time to research the podcast format make the XML file or even spend the time to get this working correctly so let's see if Auto GPT can do this for me so I have no idea this will work but I've tried to outline the problem as best I can see it I want it to automatically scan my audio folder and build an iTunes compatible XML file that can then be hosted online I also wanted to create a file called podcast.pi that I can run for myself this is pretty cool the first thing it's doing is looking for an extension library and once again I'm going to do y and probably give it 10 chances to get things working so this is interesting though it is actually trying to do things with the file system now the first it's first go podgen.pi hasn't worked so it's going to try something different this is interesting it's having some trouble installing the podgen library via pep and that's perfectly normal because Library installation and Rapport is best achieved with the package manager of course it doesn't know it's running on record it doesn't know if it's a package manager available to it and this is why you need to massage them so let's try ring that again and continuing with it but this time we'll try and prompt it slightly differently we'll continue with this I won't decide to try and use Pip this time we'll tell it not to so see here I can give it some feedback and in this case it was getting stuck on that podgen Library so let's see what it does when I tell it to you something different you see it's getting stuck in the loop this is one of the common problems with these at the moment it's sort of worked out what it needs to do but it's having some difficulty and it's struggling with that so I'm going to give it some feedback again so you get it's getting a bit confused here it's trying to see if fijian's installed and it's not actually doing that okay it's getting stuck in a bit of a loop there but you can see the potential of this if I can talk it out of that loop I can get it to actually set everything up for me and build everything which would be pretty cool but let's try something a little bit different I've heard a lot about Tailwind CSS recently and let's see if it can justify to me if I should be using it or not so I'm gonna give it 10 chances to get it right and loop through because what I've asked it do is find out what tailwind.css is compared to similar products give me a list of pros and cons and tell me whether it's worth me changing over to use it or not and you can see that it is a reasonable job of its first stab of finding out about what it is now it's going to take a look and browse some websites to find out a little bit about it and you can see the cool thing about it is it's got a big website that was bigger than the amount of tokens that we could use with GPT and it's just used chunking to summarize it in different parts so it's got that understanding and this is why these autonomous agents are so good it is using and working around the limitations of Our Lives to get them to actually think about things and basically chain together thought which is absolutely amazing to see in practice so you'll see here that it's been that it's been writing the pros and cons to a file take a look at our outputs folder for that you see we leave it working in the background we can see already it's generated this document for me pros of using pros and cons so we've got that which is great because I've got an executive summary basically of the research it's done so far oh Has It produced a markdown table for me it has let's open that in a preview and see wow that's pretty cool it's done that that's pretty cool indeed let's see where it is oh this is interesting so it's actually asked the llm to justify where summary was accurate it's a nice way of checking your work isn't it I'm just going to ask it to give me a decision because we're not going forever but there we go well that was a pretty cool example actually and it could have gone on for a while but this is the beauty of an autonomous agent I'm not spending that time myself I could give it a hundred chances to go through and get the data and go and do something else in fact if I turn my report always on I could just close it and come back to it a different time and see what was happening it is amazing how much work these things can do and automate for you and in fact we are at the very Cutting Edge of what we can do with them in fact most things you attempt with it will probably be the first time anyone's even attempted that in fact when I was using this yesterday to compare different GCSE courses in the UK it is quite likely that it's the first time anyone's done it to compare academic courses ever these things are a week old it's amazing why don't you get onto replit Fork that template pop in your open API key and see what an autonomous agent can do for you don't forget one of the really amazing things about replit is you can do all these from the mobile app on your phone in fact that's what I'm going to do next because if there's one thing that autonomous agents is good for is research and as you'll see I need to do a serious bit of research whilst I'm getting back to my vacation see you next time guys

Channel: Replit

Views: 21,141

Keywords:

Id: f7KRkNsoPts

Length: 11min 47sec (707 seconds)

Published: Thu Apr 20 2023