How To Install GPT-Engineer👨💻- Build ENTIRE Apps With One Prompt! (Quick Setup Guide)
Vložit
- čas přidán 13. 06. 2023
- Update: you no longer need the requirements.txt file. You install it simply by running “pip install gpt-engineer”. Also, the example files and your output will be found here: projects/my-new-project/workspace
In this video, I show you how to install and use GPT Engineer, which is a new project on the top of the GitHub trending repos that allows you to enter a prompt and build an entire codebase project easily. It's very impressive and already better than Smol Developer.
Enjoy :)
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Links:
GPT Engineer Repo - github.com/AntonOsika/gpt-eng... - Věda a technologie
on windows you need to use the 'set' instead of the 'export' command for the API key
Thanks, that one really helped! 😘
Thank you!!
i was just wondering why we all of a sudden where in zsh like it was expected. thanks. Edit: also changed to "gpt-3.5-turbo-16k-0613" since I don't have 'gpt-4' available. It does... aright. not sure ymmv
ooohhh thx was getting error
@@erdemtemiz6072 you have to execute the command each time you restart your pc. youre essentialy creating a temporary environment variable. doesnt have to be done via the command if you set it up permanently through the settings of your pc
Also a great walk through on how to use many simple things such cloning a Github project, good Python/Conda usage, simple, down to earth English - well done, really liked this video.
Thank you, Nathan!
Hey Matt thank you for sharing all your videos.. Ive been following you for a week or 2 now and you make some simple easy to follow videos that we all love. Keep up the good work my bro...
You're one of the few AI reviewers I watch immediately and this is why. Just stellar projects and amazing command of the techno-sphere. Thank you.
Awesome video. Short, to the point and no detail sacrificed. Cheers
Nice video Matthew! I love the fact that you explain things in a way that a non coder can understand it. Cheers!
Thank you! I want everyone to use these tools :)
The author is also a non coder.
@@MLesp Anyone that has a "100% Python" git repos is a non coder lol
@@MLesp the author of gpt engineer?
@@puppergump4117 not him the youtuber.
Once again, a mind-blowing video about a mind-blowing new tool. I'm having a hard time keeping up with all your tuts, but that's a good problem to have. Props to you!
Thank you!
Great tutorial, thanks! After a bunch of iteration I was able to get a pretty good working version of breakout going!
DAMN IT!! I was all excited to bring up "Smol Developer" because I hadnt seen you do a video about it... And then make a video about a model that does something similar, but way better... Great job man. Thank you for the video.
Awesome Matthew! Doing the research and saving us time. Love it.
Preach it man. I worked web hosting support about 5 years, and I don't know how many tickets I dealt with where people had used pip on the wrong python version. You saved some people some hassle with that info.
Took me a while to understand Python versioning lol
When I use the “ python -m pip install -r requirements.txt “ command it says error number 2 file doesn’t exist in this directory. Why would that be happening? Do I need to create the requirements.txt file manually if this is the case?
I just never saw him create the requirements.txt file manually so I don’t understand what I’m doing wrong
Freaking sweet. Keep up the good work I watch your content constantly because it is the most valuable of the many trying to share AI news
Absolutely outstanding! Thanks for sharing such amazing content!
Awesome stuff Matt! It's like Christmas everyday.
Good video! It should probably be noted to people that beyond your data getting sent to OpenAI (which is expected, obviously) this app also phones home. It seems to send all of your interactions to something called RudderStack Analytics with a tracking code specific to the author of the Git repo. Someone concerned with privacy or IP theft may want to exercise a little caution
I love these updates and tutorials, thank you
Great video Matthew I am definitely going to try this. Thanks!
Thank you for the video, I had to modify mine a bit but I did get it working.
Relevant information. Worked like a charm . Thank u
Excellent demonstration. Thank you
Great tutorial. I really like this new project. Thanks for making the vid👍
I can't believe I got this working. Took me all day. Everything from having to switch terminal over to Bash because powershell wouldn't export to the directories being all screwy. after all was said and done, it was very gratifying to actually have a running snake game. Holy crap! Now for the real fun. TY for your awesome content.
Hi. What is terminal?
@@davyprendergast82 Terminal is like Windows cmd in VSCode. I have the option of several command lines. I am on a Windows machine.
@@garyjurman8709 Sorry to ask another dumb question, but what is VScode? I just downloaded Conda but then in the vid he said go to terminal. Does that mean open a command window? But from within which folder?
That is awesome! Thanks for sharing!
This is an excellent demonstration of GPT Engineer! The ability to enter a prompt and build an entire codebase project easily is truly impressive. It's exciting to see the advancements being made in this field, and GPT Engineer seems to have taken it to the next level.
Great Video! Thanks for bringing this project to our attention (and I thought the 16k token news was the pinnacle for this week 😅)
Inspiring! Great project! 👍
You can use the pip command directly. It is the Conda environment that makes sure you use the correct version. Always.
Great video!
I appreciate you sharing your knowledge with us.
I can get the same result without using Conda ?
Bruh you are right on the bleeding edge of all of this and i'm on your coattails thanks!
That is so cool!, thanks for sharing!
awesome. we need this kinds of videos
Excellent. Thank you!
Matthew you are awesome and your content is great 💯💯 Thanks to share your knowledge with us 🥳 Keep doing this great work, you rock man 🤘 Plus +1 subs.
Oh wow, this is awesome!
Nice, Thanks for sharing this ❤
Absolutely incredible 🤩
Thanks :)
I appreciate your fluidity and tonality. So well spoken and definitely helped me understand how to install and run everything. That being said I tried installing the gpt-engineer git clone before watching your video and failed to add the conda file in my terminal. My Terminal (MAC) isn't letting me move forward with the install. Hope to figure this out before I go crazy. lol
I am having the same issues, sucks nobody helped you yet
Super helpful video 👍
nice!! yea!! very helpful & promising!! thanks & good luck!
When the brackets are around a letter, that is the default. So, if it is "Proceed ([y]/n)?" you do not need to type in "y". You can just hit enter.
Ohhhhhh. That’s a great tip thanks.
A good prompt engineer knows to use openAPI to optimize and automate the creation of their prompts if they need a little push LOL but using chatGPT to further prime in this project is a total no-brainer! I love it, downloading now! also... SUBED!!
First off note that this uses GPT-4 by default, so costs can add up quickly.
I tried to get it to implement chess. While it output a good set of files and classes it left much of the implementation as "# Add your code to implement specific movement rules for Pawn" etc. Just a comment without code. I then fed that generated code back into the system as part of `main_prompt` asking it to do the implementation and it failed to do so, outputting "# [IMPLEMENTED CODE]", which was kind of funny.
lol did you get trolled by a chatbot?.. 🤣
It's a good idea to include in the prompt something like:
"Please note, I would like you to complete all of the code i.e Do not include #Add your code to implement..... blocks, make sure these are fleshed out"
Good find. Thank you.
I am so happy you at 2:32 also tell how to create this missing requirements.txt file. For those who did not see it, there is a pip freeze > requirements
If you are running on Windows, you will get an error when you try to use the EXPORT command. Use SET instead :)
It only works if you have access to the GPT-4 APIs
Change model in def chat() model: str = "gpt-3.5-turbo",
@@CesarVegaL yeah I changed it but gpt3.5 isn't really able to make it done, that's why I wrote It only works if you have access to the GPT-4 APIs.
Save your time and don't try it with anything but gpt-4 :)
🤩 this is amazing stuff
I don't know python but your vids are inspiring - keep it coming :)
Thanks :)
Thank you for this video
It is crucial that we provide support for these types of projects to give them the encouragement they need and help them progress further..
Wow, thanks for sharing
Amazing!! Thank you!😀
No problem 😊
I noticed this one when it was first posted on twitter. Kinda thought about trying it but I'm spend-shy. I spotted your video, being really lazy about trying anything else. I threw the first thing that entered my head at it. A graffiti site where you post any message you want. (I need to add a rotate feature :) it took some work with gpt4 but it worked. It's exactly what I asked for. Saves the graffiti to the database. Loads when you load the page. :D The author of the github deserves some kudos.. that reminds me I should go star their repo.
Update, kinda works on small projects. Anything big falls apart. Gpt decides to write two functions and calls it quits :D
Coding AI is getting better by the day
Haha, I love this. I literally asked if the snake game could be created with SMOL and then this came out.
Great Video! This result is really amazing. I do like that this model asks for clarification of information! That has been missing from many of the models. Also that this was actually able to make the "Snake" game that worked is just as you said Mind went Boom! hehe....
I don't quite understand how this model doesn't have pre-existing code iteration... One possible work around... Could your prompt simply be something like "Improve my program called "MyProg" and then when it requests more info, paste in the program?
Woah bro amazing
Hey, Matt! Love and share all your videos, I am interested in making a top-down game in Unity using AI for most of the leg work as I am an IT analyst but not a game dev... what tool would you recommend for helping this type of project? Thanks!
So well explained thanks very much . Greetings from James J in Limerick city Ireland 🇮🇪
You the man now dog!
Truly amazing and I wander how far you can go with this! Thanks for sharing this very nice tutorial! 2 question pop up; it’s writing all different classes, would GPT Engineer be able to write complete code using for an App using, React, CSS, JavaScript and let’s drop in SQL or SQLite? And my second, I’ve seen it giving small parts or call it suggestions of SQL5 code, but would Engineer write the whole code in SQL5 as well? Thanks for your incredible nice work!
Great video again! Very excited to see real world apps fully built with ai
Unlikely.
Well done thanks
Very cool, it'll be even more exciting when it doesn't require chat-gpt & can use local & open source modes instead.
I wonder if we can do like a simulation between many agents like langchain allows and simulate a whole scrum team. Basically, set some requirements and then check the result. When the result is out then inspect the result and provide some feedback, if what already is coded is in memory then we could do a sprint in 5 mins maybe 😀
I’ve been wanting to build this myself. 3 agents: 1 to write code, 1 to review code, 1 to write specs
@@matthew_berman They may beat each other up on the code review :D
@@matthew_berman Hmm sometimes if there are too many agents the agents could get into an infinite loop. The only way out of the loop is maybe to reduce temperature, introduce some randomness, or something. Do you have a repo that you are working on?
@@aoeu256 not yet. But I plan on building this in the coming weeks.
@@matthew_berman Don't have the repo on hand but I feel like I saw something like that a week or three ago. It's all a blur now trying to keep up lol. Might have been on twitter. .. not really a helpful comment but I thought I'd mention it in case it pops up from a quick search.
Amazing video. Thanks for the simple and quick explanation. I'm keen to see if this also handles UI aspects of development. Or is there another repo better for this?
Yes it does, the secret is to use GPT-4 to generate a detailed, high quality prompt
This is really excellent, very concise, and to the point. I presume it also runs on MAC OS environment?
I think if you made a simple stack overflow crawler you would get the same result from it. It's such an overused trivial task. It would be more representative if you tried to code something that doesn't exist yet. If you need some ideas to test future stuff on, I can send you some
You mean something programmers already do?
I thought I was subscribed to more than enough channels on AI, close to a dozen.....but now I just added another 👍
i know i asked before but id still like to see a video on how to install any new or old models that have the ability to run on GPT4LL i love all your videos keep it up !
Thanks!
i wish this would use a local model instead of the API. but still cool to see. tnx for sharing.
Great video. I went through everything and discovered one edge case for using the code base. The code requires a GPT-4 API key. Most users have access to only the GPT 3.5 Turbo API keys. To get access to the GPT-4 API keys, you have to request early access. The code will display a traceback immediately upon running the software and will say that 'gpt-4' does not exist even with a valid 3.5 API entered. Needless to say, I signed-up. Fingers crossed.
liked and subscribed, Cheers.
At 4:06, I keep getting the following error (in Windows) - "python: can't open file 'C:\\PROJECTS\\AI Lab\\GPT-Engineer\\gpt-engineer-main\\main.py': [Errno 2] No such file or directory" Does you have any idea of what I'm doing wrong? Thanks again for this amazing video!
Now, imagine A third AI that uses GPT-Engineer and ChatGPT to actually automate the "clarifying questions" bit of the workflow.
how can we connect that to something like spreadsheets on google
have you tried auto-gpt yet with self feedback? should be able to integrate gpt-engineer..
@@edeltobi ws thinking exactly that
So very cool, it will never be as hard/bad as it is to develop as it is right now.
Yes. And engineers will become more productive with AI, not replaced by AI!
@@matthew_berman Well, if a engineer lead usually had 3 junior developers write unit tests, he can now just do it with 1 text prompt in a tiny fraction of the time, so I think it may be not great for junior developers right now
It just means the bar for required knowledge is raised. The many "software engineers" that code buttons and UI for web apps will now have to actually learn software engineering principles and system design
@@MCroppered interesting take. For all intents and purposes I think you are completely mistaken. Autogpt was coded with chat gpt in 3 days by someone with zero coding experience. As these tools improve, the pre requisite experience requirements are being eliminated.
@@x1expert1x you're mistaken. generative AI is the great equalizer when it comes to coding. junior+AI is roughly the same as senior+AI, with seniors having a slight advantage due to sheer experience (they're better at prompting), but that gap is closing fast and so is the paygap between the two
idk if you know this, but at 3:03, you can run “code .” and it will open vscode in the directory you have open in the terminal.
Thank you!
@@matthew_berman since you using the insider's version, it's "code-insiders ."
thanks Matt, you are the man for sharing this. I've been getting a bit frustrated with smol developer. and tbh I'm getting A LOT frustrated with still having to wait for GPT-4 API access. I can't get experience the full capabilities of any of these tools being developed without it and I've been on the waitlist forever. Any tips on getting access?
Maybe the author can use embedding sao it can read the code from a vector database. Amazing video btw
Love your videos! It is too bad that you need a GPT-4 API Key, still on the waitlist :(
Mind blown x2! ;)
Matthew teaches so much gpt I'm beginning to wonder if the headshot recording is AI'd as well XD
You probably won't see this but your videos are some of the best and most consistent quality content helping me to keep up with everything changing so fast. So thanks for all your work.
And we're just at the early stages. Just imagine what it would be like next year
Love the video, not knowing anything about coding, still feels like I could do this.
^^^ with that in mind, what or where would be a good place to start learning the basics of coda* and python?
Trying to make best use of this content!
Ask Chatgpt to teach you!
thanks!
Top CZcams channel I follow these days 🙌 I’d like to find a tool that actually executes the Python code and resolves errors as it encounters them, any recommendations?
You need to add the caveat that this is for GPT4. Not many of us have that API access. Would be nice to know before setting up everything. Otherwise great information on cutting edge projects.
at this point you can kind of safely assume the vast majority of actually noteworthy projects use gpt4. Fun exercise is to fork the repo and try to make it work with 3.5 turbo lol
@@markpfeffer7487 did u make it work with gpt-3.5-turbo? I'm trying here, kind lost actually
Do you require premium to access the api ?
@@ulissesmalanski514 been at work, haven't tried. I would be willing to bet it would take too much work and iterative prompting to make it worthwhile. Id wait for a wider gpt 4 api release for this one lol. Simpler tools that use gpt4 can be adaptable. Especially now that system messages actually kind of stick with 3.5
@@ulissesmalanski514 Change model in def chat() model: str = "gpt-3.5-turbo",
Works for trivial apps where you basically will not separate from the masses of other apps. And design is something it doesn’t at all.
Thanks you subscribed and liked
Thanks!
damn!! lets try that right now to create a saas. (thanks )
Can't even imagine where this will be in 10 years.
great video
Thank you :)
App Digital Vending machines incoming 2024. Thanks for keeping the vid short bro.
You got it!
Nice video. Appreciate the python version tip.
Hi Matt. This looks amazing. Is this for Mac or Windows? I'm very new to all of this :)
Thank you for this video. Btw, how do you have your cursor set up that way?
Sup Matt, love your videos. Did you use GPT-4 or GPT-3.5 Turbo?
Great job Matthew, As I don't have access to GPT-4, I have changed the model to gpt-3.5-turbo. It generates the code, but leaves everything in an “all_ouput.txt” file, nothing else. Then it's time to create each file according to what that document indicates, some adjustments and voilà, the game works fine.
This should pinned too! It helped meee thank you!
Can you post the steps you took to change Model plz, I changed in "def chat (model: str = ""gpt-3.5-turbo")" and got error [ openai.error.InvalidRequestError: The model: 'gpt-4' does not exist.
Do you need to have premium to access the api ?
@@stef4614 You don't need it because you define the model as gpt-3.5-turbo that is free
@@stef4614 if you mean do I have a pay as I go Acc, yes
Hello, thanks for the content, love that. But do I understand correctly, that OpenAI is not going to give any free credits to new users anymore? If yes, do you know if is there any way to modify this program so that it uses some other model (for example, calls some Hugging Face Inference API endpoint or some model that I`ve hosted) to build projects instead of GPT?
Can you show us the logs of the single prompt you used to achieve that outcome and what the response was?
I think it only works if you have access to the API for gpt4 , I just got in the wait list I guess just need to wait now. Amazing video btw.