AUTO-GPT: What can it really do?

yo what's going on so I'm recording this video basically to document some exploration of Auto GPT um my intention here with this is to just dive into the files kind of try to understand exactly how the software works as of today and ultimately get it to work at an acceptable quality because so far as I've played with it like it works but how do I actually dive into using all the commands and how do I dive into actually getting this thing to Output what I want it to Output so essentially I'm going to go through this I'm going to record it um and then hopefully someone gets to watch it and they can learn from this and at the same time uh you know they can also uh just kind of go through it with me basically so I would like to get to know how to use this tool so that I can help others get to know how to use this tool because it's a new tool it's if you don't know about it it's called Auto GPT about a month ago it got a good amount of hype um it's an open source code base that allows you to plug into open AIS API so you can use their GPT models with this you can use GPT 3.5 and GPT 4 and essentially it's a good amount of python code and it is supposed to um be an AI agent or at least that's what they said it was supposed to be so the other thing to acknowledge about Auto GPT is that it's an experiment and the creator has open sourced it and a lot of people have jumped on to developing it and now it's really taking a life of its own and there's a lot of real-time feedback and there's a lot of devs from across the world that are contributing to the code base to improve the product and ultimately right now pretty much in order to use it you have to have some kind of um code skills you need to be familiar with python at least at a beginner level um they are planning to release a user interface a graphic user interface and you can get on the wait list for that now as of May 1st at least but in the meantime um you know I think it's pretty cool place to learn also and do some real-time open source product building with other developers and jumping into their Discord it seems like there's a good amount of community being built in there and personalities that are in there by the way uh I haven't been recording videos in a long time so the funny thing to me is that the first video that I come back to record I'm rocking a mustache a little backstory on that I just wanted to go see Super Mario Brothers with my youngest this weekend and uh it was a great movie I loved it he thought it was awesome the graphics were awesome computer Graphics were awesome sound was awesome storyline was awesome and uh you know I came out of it and I was like yo I want to be super I want to be Mario so I'm rocking the Mario uh mustache but anyways um yeah so I'm gonna dive into it here and hopefully uh you know learn something so I'm gonna go ahead and start with the auto GPT folder and basically this is the software folder okay so in here you have your python files and these files are what run the program so let's see let's go into this one init.pi nothing there yet main dot pi yeah there's still some things that I myself I am investigating um I haven't used Docker yet so this is a good kind of project for me to get familiar with it and let me see I think a lot of this is built for Docker to be honest what I'm going to be running it locally for now until I get up to speed on docker commands so this is important in here commands commands commands okay so podcast what we got here anything nothing yet code evaluation from locations a function that takes in a string and returns a response from create check completion API call parameters code to be evaluated evaluated where it turns a string from create a check completion a list of suggestions to improve the code function equals string okay so I guess this is the function for passing in code to analyze audio from file yeah I think this is really cool and it's really interesting to me the whole that it reads back to you changing the kind of the interface of how an end user interfaces with this API or this AI software so that's pretty cool and I have something in mind in the back of my head that I want to build that would incorporate a mainly You Know audio format interface cool yeah here's more docker so yeah I still have to get familiar with Docker because I haven't played around with it but let's see this is what I like about this project is some of the problems that are still open to solve that's cool cool yeah opportunities to learn get better at Building Products is pretty much my main motivation here second motivation is learning how to use the product as is and then teaching others how to use it because I think that something like this is the beginning it's the very beginning of interesting not seen before products because right now you're seeing a lot of CopyCat kind of products come to Market and a lot of them I mean this is just part of the hype cycle a lot of them you know I shouldn't even call it the hype cycle it's like the Innovation cycle a lot of these products they're just copycats of other products and already through this Innovation you have seen uh how a good amount of them have become obsolete relatively quickly but I think that it's good to put Capital to work because as capital goes to work and these lessons are being learned um you know other teams benefit from those lessons and you even you know new products come to mind and pivots could happen and you know with capital working Innovation gets to come to the surface and that's just a benefit to end users and customers and Society so yeah it's cool anyways so let me go back up here this looks like cool so these are some functions aiming to do further automation yeah so this is interesting too so there's some functions here that read files and ingest files which is great so that the llm which stands for large language model can go ahead and parse through that understand it and try to make sense of it in terms of outputting a prompt um and in this case a next action this is important because it starts to create a starts to create context to what you're asking it to solve cool Google search dot Pi this is interesting Google Search Command you turn the results of a Google search that's a short very short function oh my bad my computer just was freezing up Google search return there's also the Google search query increase uh you know on a daily basis which is cool not for everything but for specific things right now I'm thinking that research is probably the best application for this pretty basic image generation so here you can looks like you can use Dolly hugging face sdwebui to generate image cool I do need to dive into hugging face more generate an image with dolly yeah again like you know this could potentially with a set of prompts generate content in the form of an email with images or a blog post or even website content honestly like uh more static and uh generate image stable diffusion yeah so what's cool about Auto GPT is as a quote-unquote agent you know it's it's plugging into different AIS through their apis and bringing them together to help create things automatically faster which I think is cool from the future for a function that takes in code and suggests and returns a response from create chat completion API call that's cool um basically like a coding assistant this function helps you uh with that look at date and time that might be important oh this is new the Twitter dot pi okay yeah so this is something I've been seeing is that I think this is like a great use case for something like this is it's just like a it's basically a Twitter bot on steroids this is plugging into the API and as of right now you know the Twitter API has changed so before it would have been even better where you could use Twitter's endpoints to ingest conversations from Twitter into this because I think that would be really powerful in terms of real-time insights but now it looks like basically hey maybe that's something to expand on actually like that's a thought you know to to maybe build a plug-in for this that allows you to ingest some of that data that might be cool and get some insights but anyways uh yeah this basically allows you to post on Twitter using their API playwright okay interesting okay beautiful soup so this is for web scraping essentially which is great more or less it allows the agent to go out using Google and read web pages ingest that uh and then draw new insights and allows it to reprompt itself from what it's already learned so that's pretty cool so yeah again like you know even with something like this like uh you kind of need to have some basic understanding of some of these other python packages to truly fully Leverage uh Auto GPT for right now um which again is why I wanted to explore this and really just learn as much as I can on how to use it so that as you know I come I think up of problems or potential improvements you know I can help Implement them but also just jumping into the Discord and sharing those ideas as well as I'm using this thing uh and going through the code base and then hopefully getting to a point where I'm pretty proficient at using it and you know can help others become proficient using it as well now they are talking about releasing their releasing their GUI which is their graphic user interface just basically a front end for non-technical users to be able to use the tool but even with that I still think for right now being able to use it from the back end without having that kind of wall that's like what's going on is really powerful in terms of really understanding the ins and outs of how to use this thing because who knows maybe someday you know you don't have as much visibility to how the product actually works and imagine being able to use uh Instagram as it was evolving and being built you would be a power user and you would be able to grow audience and build Brands uh relatively quickly and even help others do that so I don't know it seems like a really good opportunity um okay cool yeah so for now I think that's pretty interesting uh uh maybe super boring watching me do this but again uh I've already kind of seen that there are some additional uh topics to explore right in terms of using this also another thing to to note is that this is the auto GPT version 0.2.2 uh and on the repository there's a master Branch that's I guess being worked on relatively uh frequently and sometimes this thing breaks so you have to like use it while it works and then configure it to fully use it and then if it breaks you have to figure out why it broke and how to reconfigure it so yeah it's definitely time consuming and it's a hassle for sure but um that's why I'm taking it up so for now I'm gonna end it there uh I definitely I definitely feel like there's a couple of there's definitely just by going through this there is definitely some additional learnings that come to mind right so you know even just like you like for example with selenium and the Chrome uh you know automated web browsing there's some configuration that needs to be done there in order to fully leverage it here so things like that I'll probably break most of it down right now I'm just recording I'll probably break most of this down uh video edit it and then um yeah I'm excited to kind of see like how this comes out so cool thanks for being here I'll see you soon

Channel: Alexander Eick

Views: 648

Keywords:

Id: UHqZ61pf6SY

Length: 21min 22sec (1282 seconds)

Published: Mon May 15 2023