GPT Engineer: The Future of Coding?

--- type: transcript date: 2023-06-20 youtube_id: r6n1ZC3nRQA --- # Transcript: GPT Engineer: The Future of Coding? all right in this video I'm going to be showing you GPT engineer which is a new open source project that allows you to build out full-fledged applications from Simply writing a prompt so it will go through and create everything that you need in the directory and the dependencies install everything right test for you and give you the option to execute it so it's a pretty interesting project and pretty amazing the different features that are already built into this so to go ahead and install I won't run through step by step how to do this just go to the GitHub repository run through the steps if you run any into any issues with these installation steps just take those errors put them in something like chat GPT or do a little bit of Googling because usually with these types of things if there are issues with dependencies and whatnot they're sort of variable to everyone's machine so once you work through these steps hopefully it's just a seamless uh process you have python install the recent version all of that good stuff you get through this smoothly you'll have to go ahead and get an API key from the openai website so just go over to the platform.openai.com account slash API keys and then you can simply create a new secret key so once you have that you can go into the root of your project and you can simply within your terminal export the API key for it to use so the one thing to note is you do need gpt4 access so I do have gpt4 access I haven't tried this with any other model so if you don't have gpd4 access I actually can't say what this does or if it works at all so just one thing to note with that if you run into any issues you might have to request access to gpt4 at this stage so the next thing is we're simply going to be taking the example folder and I'll show you here we're going to be taking the example folder which has a simple prompt and then we're just going to be copying that over and I'm going to follow along like it shows within the documentation here but just add a couple things so this is going to create the new directory and template here as you see but if you want to have you know a custom name for it you can go ahead and do that just make sure in the subsequent steps you run the command with that new name that you put for it so once you have that we have that same template here we can keep our example template nice and clean we can go ahead and copy this over which will run in just a second so this is going to be how we actually execute the scripts uh within GPT engineer on our prompt here so if I just say I want a simple let's just say game I'll be sort of ambiguous at Parts in HTML and let's just say HTML CSS and JavaScript only so let's try a simple example and let's run through this here so while this is booting up I'll just touch on briefly there are some examples here if you're looking to just test something within the application different than the main prompt there you can look at Benchmark there's a handful of different examples here that you can try out so now within here you'll see that it's giving the program a description so it's elaborating on what we specified it's going to say okay the game is going to have a start screen the player is going to control the character the moves left and right left and right avoids falling obstacles so it's sort of creative right off the bat it's creating um a game without us specifying what that game is now the thing to note is obviously within the main prompt you can be a lot more verbose and descriptive if say you have a particular thing that you want to build you could go ahead and put in a ton more detail so once it's sort of specified within natural language the features of the application that it needs it's going to go ahead and start to get a bit more technical so it's going to just say okay here's the core classes the functions and methods so it's going to first specify what the files are it's going to specify what are in each files just sort of broadly and then within the JavaScript you'll see that there are the different methods that it's deciding that it will need for the various functions for the game to work so the other thing too with this is it will create tests for your application and one thing to note is with the quote-unquote identity of your GPT engineer if there's certain things that you'd like to tweak or change or remove you could go ahead and go within the identity and change some things here within these steps so this is something that this is obviously based on the the core repository here but depending on your needs or say you're running into difficulty at one step where it's not doing something quite as you'd like you can go ahead and update the identity here so you you see here it's generating a package Json now if it was an application that required a back end that pack could Chase on could very well have different dependencies so say if you're trying to set up a server it might go ahead and say okay you're going to have a dependency of Express or whatever you're using here and one thing to note with back-end examples is once you're at the end it will go ahead and prompt you for whether you'd like to install and run the code so you'll see here it's going to have a main.test.js it's going to specify a handful of tests here okay so you see it's describing different things I'll just look back here so it is going to be using just to test everything and we just run down here now the one thing to note is you see this this goes pretty quick so you sort of have to think of this as you're giving a job to an employee or you know someone you work with or something and you're just handing it to them and you're not necessarily meant to sort of watch all the steps like I am right now obviously this is different to sort of demonstrate it and show you what it is but generally you can sort of think of it as if you're handing off work and you're going to get it I'd say the end of the day or whatever and then you'll assess okay is this right or what needs to be done or what bugs might there be Etc so that's sort of conceptually how I think about these tools anyway okay so you see it's got the HTML it's got the CSS it's got the JavaScript logic and it's still working through some of it by the looks of it so within the project we can go ahead and look at a couple things as it's running through so you'll have the logs you'll have the specification so you can see the specification uh here of okay within this is going to be the simple HTML CSS JavaScript it's going to have the unit tests like we saw here and then the logs so there's going to actually be a Json object here where it's going to show you the prompts with the openai API so you can sort of see the conversation back and forth of what's being sent back and forth to the llm and gpt4 model okay so you can sort of look through the logs they're sort of helpful to see initially what's going on but ultimately what we're going to be using and looking at is our workspace here so this is everything that was generated so you'll see we have our NX HTML we have our main JS we have our tests we have our package to install just we have uh all sorts of stuff here we have our styling readme and then once it's done executing here you can simply press enter or say no to run it so let's just see what happens here so it's executing the code now within the context of a front-end application it might not actually need to execute much maybe it needs to install the Json now it didn't exactly do that but let's just see here so I'm not too concerned with the tests for this I'm more curious to see it but just in case you run into this and you have to manually go in and run anything so you can just list out the directories here you can go into projects you can go into your new projects you can go into workspace now within workspace say you want to install the packages so in this case we want to install just we saw that we clicked enter and it didn't actually run we can always do things manually like this as well thank you so while that's loading we can just go ahead and let's just open this within live server in a browser tab here and we can say okay here's a simple game use the arrow keys to move left and right to avoid falling obstacles okay so you see a game it's sort of like asteroids it built for us now let's see what if you actually go get you see you get hit it's a game right so all I did was specify that I want a game and it successfully built me a game now the one thing to note with this this is a working example and this is great to show sometimes the prompts that you put in especially if it's a more involved application there is um you know as the complexity grows the likelihood of Errors coming up obviously increases so keep that in mind but definitely a very cool project to check out very impressive stuff by the contributors really excellent work I'm really excited to track this project and see where it goes I think it's a conceptually spot on with what you can do or what we'd like to do with some of these llm tools as coders simply specify the you know the feature set or the application itself and it to give us you know a boilerplate to work with so uh the other thing to note is um you do have the identity that you can change here so I would be curious to see uh if there's anyone within the comments if you've changed anything here if you've refined anything that have given you better results whether it's slightly better or markedly better let me know but if you found this video useful please like comment share and subscribe and otherwise until the next one