AutoGPT Test and My AI Agents Effortless Programming - INSANE Progress!

today I want to show you how my AI agents take python coding to a whole new level when they cooperate to create small mini games and apps with a UI in only two minutes I also want to show you my auto GPT project where I try to create an AI Agent form with the ultimate goal of creating an AI system so let's just get started first let me just show you a quick flow shot here of the system so we got a python Hub in the middle we got two Chef GPT agents that are hooked up to chatibility 3.5 API the system starts with a trigger message that's going to start a simulation between our two sharp GPT agents they have both access to Google so they can go outside the frame and search for info bring it back and tell the other agent about it their goal in this system is to work together and to create a working python code of our description let's say we want a game then they're gonna cooperate try to make that game and put out a working python code so that's it let's dive deeper here we are at our python Hub I'm not gonna go through all of this code now that's just gonna take too long but we are gonna look at our agents so we have agent 007 the role is you are a professional software developer with special knowledge about Python and games keep your messages short focus on the code your task is to create a text box to write in with a UI in Python with agent 69 your task is also to give critique feedback to agent 69 Converse in this structure response reflection code and critique we have the mirrored instructions for 69 but he is gonna critique agent 007 right I had good results with doing this mirrored approach but I also want to experiment with keeping that both the system roles quite different so they can work off each other's strength that's going to be for another video now let's try to create this text box to write in with the UI okay so let's run this now I brought up a timer here so we can get an Insight of how long this takes so let me start that over here enter and go okay so we start off with the trigger message right so hello 007 I'm 1869 I'm here to help you create a simple textbook in Python we can use the Thinker Library which allows us to create a GUI easily and here you see his reflection it's important to make sure that we are addressing the correct person to use a professional tone in our conversation and then he puts out an ID for a code this code or the code creates a simple text box using the tincter library which is a good starting point we don't have any labels or buttons it could be improved okay so agents zero come back and says thank you for creating the window it's important to build up the previous code and address the limitations in a user-friendly manner so he looks at the code comes up with some IDs it now has more function abilities which makes us more user friendly the previous limitation were resolved and then it comes back to agent 69 again trying to it reflects on providing constructive feedback and suggestions for improvements can lead to a better outcome that's his overall reflection right and they go back and forward try to make good ideas for the code this is the code we end up with we are gonna test it so let's copy this and paste it in and test it okay so here we go here is the result write something okay so I guess it's working you wrote error text textbook is empty okay so we got a message all about AI submit it's not saving or anything it's a very simple UI yeah good job now let's try to create a whole bunch of different things and see if it works now let's try to create a tic-tac-toe game with a UI in Python okay sounds good over here start the timer and run okay so that was about 2 minutes and 50 seconds let me copy that and test it okay let me enter all right this looks good so we got a code here okay that was a bit strange let me see if it works X ah good yeah okay so what happened now okay so it stopped but uh can we restart it ah here we go X wins I wonder why that didn't happen earlier okay cool that works let's do something else okay so now let's try to create an app that checks the weather in Norway with a basic UI okay let's run that start the timer and go okay so that was about one minute and 20 seconds let me just grab this and see exactly what this is okay so let's test this now uh welcome to the Norwegian this is not exactly a UI but okay welcome to Innovation weather app check uh please enter name some cities in Norway okay let's do Oslo and Bergen wow that works it's not that sophisticated but at least it works uh I just fetched a free API on the web and yeah current temperature in Oslo 10 degrees broken clouds 13 degrees in Bergen a few clouds so we can just do a quick UI on this if we wanted to I'm not going to do that now let's try something different so now let's try to give them some Freedom here so we can create a new strange simple game with a UI in Python so I don't know what this is gonna be but let's check it out okay so run that start the timer and go okay so that was done this is very exciting so let's just go all the way up on top here because I want to see how they taught it uh hello agent I'm glad to work on this project with you sure let's get started we're kind of a game are we building Agent Zero seems eager to start but we need to clarify what game we are building agent 69 just start by setting up a simple UI that's a good start right and I agree with your plan as for the game I was thinking we create a simple guessing game using numbers the players will have to guess the number between 1 and 100 and they give them hints if they are too high or low and the reflection is a simple guessing game could be a good start for us to work on let's start adding witches for guessing mui we need to label and enter the widgets so they gone on and on about this they try to improve it here's the updated code great catch okay so the code seems to work fine but if the player entered an invalid input like string instead of numbers it will raise a value error great catch we should add a error hand error that is very that is very good so they found a weakness and they cooperated and fixed it so remember they only spent three minutes on this so they just went back and forward tried to implement other stuff is this gonna work I don't know but let's try it so let's copy that and let's give it a spin okay so we got something here I don't know we got two windows let's expand this I don't know what kind of window this is can I grab that okay so here we got two windows so it's I hope you can see this it's a bit smaller it's basically enter your guess one to hundred let's start and we have got 10 attempts so we can choose our difficulty here it doesn't seems to work though but okay so let's do 50. your guess is too high you got nine attempts remaining too high okay let's do 30. too high 20 too high ten congratulations you guessed the right number wow so let's try hard and replay okay so I guess we still have 10 attempts so that didn't work let's do 50. it's too low 90. too high I guess it's 75 to low 80. okay so you get the point that was cool that was a good idea it's very simple of course but for three minutes that's quite impressive okay so just remember this is still with chat GPT 3.5 and that is not so good at coding as gpt4 soon I will have upgraded this to GTA4 I just need one more API and I'm ready to go that is going to be very interesting I just want to finish off this video by showing you something I created in Auto gbt so let's just head over to Auto GPT and take a look at that okay so let me show you the plan I have for this Auto GPT system so we have the AI that's called Agent Farm 9000 it's an AI designed to autonomously create and manage specialized GPT agent that can generate and reproduce other agents with the ultimate goal of developing a comprehensive AGI system so our goal number one is going to be create an agent capable of understanding the concepts principles and architecture of agile systems including the ability to create manage and reproduce specialized agents goal number three is going to be create an agent that can continuously analyze optimize and refine the replication process ensuring the resulting agent contribute effectively goal 4 is going to be an agent that can coordinate and manage interaction between various agent within the HR System enabling seamless cooperation knowledge sharing and task presentation and our final goal is to create an agent responsible for monitoring the AI AGI systems performance identifying areas of improvement and implementing unnecessary adjustments to maintain the system stability scalability and adaptability so this is quite the interesting case so I think we are just gonna run it and I'm gonna run it let's do a hundred continuous and then we will get back and take a look at where we stand okay so Auto GPT has been running for a while now I just wanted to take a look now so where we stand so it started off with this plan here to achieve the goal of creating an agile system I need to start building agents that can an agent that can say focus on this task and generate other agent so it started by creating an agent called AJ Focus so this prompt is key principle of systems right and it wrote this whole kind of list here then it went on to start an agent named AGI expert trained AGI expert to understand the concepts principles and architecture of AGI systems so yeah I started this AGI expert agent and it kind of went on to find some information about so message the AJ Focus agent start working on a report on AGI systems ensure that the agent is actively working on the task and managing process okay that's cool and I wanted to start a new GPT agent named AGI designer so design and Implement self-replicating agents with a diverse and complementary skill set enable to continuous analyze and optimize and refine the process so I really enjoy the start here I think this is what I am looking for I want to build up agents with different kind of skills right so now the thought was now I created the focus on Asia expert agents I should focus on training Asia expert to create and manage yeah I will Begin by defining core competitors specialized in GPS should process to achieve this goal so it started doing this do nothing command that's not a good sign okay so we did a Google search for creating HR systems that's good okay so it found 40 000 characters on this website let's take a quick look at it so it was this website artificial general intelligence on techtarget.com okay so that was not bad okay I wrote that to file let's take a look at that expertise in ml and AI knowledge of fields cognitive science neuroscience okay so so far so good I like this I've gotten so far before so now we created a new an AGI manager select a managed agent to fulfill the required to compute it towards the goal okay so we have a manager message the Builder now we created an AGI researcher search and browse websites for gather information on systems and synthesize it into a concise report okay that's good we want that but then we got to here and it kind of went off the rails this happens sometimes for me you can see it did the same prompt forever so this is a working progress I might have to change up the the goal here a bit I haven't really decided what I'm going to change I kind of think I need a more of an overall goal in just creating an agile system so I'm going to work on that and I'm probably gonna do a video next week or whenever you see this it might be out so yeah very interesting a lot of fun with auto GPT and I got some cool videos coming up on this baby AGI Auto GPD Trend so look out for them if you made it all the way here maybe you want to try out my system that I just show you in the beginning of the video if you want that check out the link in the description below to my membership and thank you for tuning in see you in the next one

Channel: All About AI

Views: 161,615

Keywords: autogpt, ai agent, ai agents, autogpt agent, ai automation, chatgpt, python, ai programming, autogpt github, gpt 4

Id: L6tU0bnMsh8

Length: 14min 19sec (859 seconds)

Published: Sun Apr 16 2023