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
@@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
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.
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
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
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.
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.
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"
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!
Hey, When I cloned the repo, I am not able to follow along with the steps: python gpt_engineer/main.py projects/example ─╯ Traceback (most recent call last): File "/Users/saymukhe/Personal_Code/gpt-engineer/gpt_engineer/main.py", line 10, in from gpt_engineer.ai import AI ModuleNotFoundError: No module named 'gpt_engineer'
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.
@@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?
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!
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 😀
@@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?
@@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.
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.
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
@@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 :)
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.
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.
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.
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
@@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
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.
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
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.
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.
@@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
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
User: Given that LLM like you are able to write software, soon the job of software developer will become outdated. If you were a developer what would you do given that sooner of later all the fields will became more or less automated? Gpt: As an AI language model, I don't possess personal ambitions or intentions. However, I can provide some guidance on how software developers can adapt to the evolving landscape of automation and AI technologies. 1. Embrace automation: Instead of viewing automation as a threat, embrace it as an opportunity. Learn about emerging tools and frameworks that automate repetitive tasks in software development, such as code generation, testing, and deployment. This allows you to focus on higher-level tasks that require creativity and problem-solving. 2. Cultivate unique skills: As automation takes over routine programming tasks, it becomes essential to develop skills that complement automation. Focus on areas like architecture design, algorithm optimization, data analysis, machine learning, and human-computer interaction. These skills require critical thinking, domain expertise, and creativity that are difficult to automate. 3. Keep learning and adapting: The field of technology is constantly evolving, and staying up to date with the latest advancements is crucial. Continuously learn new programming languages, frameworks, and tools. Explore emerging technologies like blockchain, virtual reality, or Internet of Things (IoT), which are likely to offer new opportunities and challenges. 4. Collaborate and specialize: As software development becomes more automated, collaboration and specialization become increasingly valuable. Engage in cross-disciplinary projects that require collaboration with experts from other domains such as design, business analysis, or data science. Specialize in a specific industry or niche, where domain knowledge is essential and automation is less likely to replace human expertise. 5. Emphasize soft skills: While technical skills are important, soft skills will continue to be highly valuable. Develop skills in communication, problem-solving, leadership, and project management. These skills are essential for effective collaboration, managing teams, and understanding user needs. 6. Adapt to new roles: The role of a software developer may evolve into new areas such as AI ethics, data privacy, or software consultancy. Stay open to exploring new roles and opportunities as the industry changes. Consider pursuing advanced degrees or certifications to enhance your expertise in specific areas. Remember, automation and AI are tools that can enhance the work of software developers, not completely replace them. By adapting and evolving alongside these technologies, software developers can continue to contribute to the industry and find meaningful roles in the changing landscape. But yeah you're fucked
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.
Given snake type games have been around for 40 years or so surely an intelligent AI would just output an existing example of the game and then translate it to whatever language you want. How do we know it’s interpreting the prompt and writing the code from scratch as opposed to pulling an example etc.?
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...
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
you still need to be the engineer understanding the problem that needs building, to be able to explain it to GPT-engineer or even ChatGPT. Never built a snake game but by just guessing, I can conclude, its relatively simple. Maybe better example would be a Pacman game or Tetris. Those would be more complex. How about a side-scroller game? :P For a toy or a helping tool to give you ideas, yes, its good, but to build a complete software solution with thousands of requirements, I doubt so.
I'm not sure this is a fair look at the functionality -- esp when you say it's the only generator that was able to build the snake game in a single pass. The reason is because it appears the dev optimized this for the snake game -- we can surmise this by the fact it's the example project. I'd love to see you do a couple of other game gens, like maybe flappy bird. Or breakout.
exactly. respect to the dev first and foremost, but there are lots of gaps left when you let the script loose on one of your own ideas. still amazing though.
Yeah as soon as you start tweeking it kind of gets lost. I'd only had 3 runs. First with out tweeks which work but created a over simple game, trying to improve the game with more prompts kind of made broken games when parts where not implemented. I'd say next step in development would be a review function that can test and fix issues and complete parts that are not written. That would require a more rigid design imo.
Common just found this channel, but if this is the type of content (trying to generate a game that is the example project...), will have to unsubscribe
For some reason after activating conda, when I try "python -m pip install -r requirements.txt" I get this error from the terminal on my macbook: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt'. Do you have a fix for this or what may be occurring? Everything was working fine up until this point.
Is anyone getting this message; ERROR: Could not open requirements file: Errno2 No such file or directory: 'requirements.txt And am using the Anaconda Prompt(minicomda) Please help
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!!
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?
when i got as far as 'python -m pip install -r requirements.txt' I had an error that it couldn't find the requirements.txt file. Where do i locate this file? I didn't see it in the repo for GPT-ENG.
I get this when trying to execute python -m pip install -r requirements.txt: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' What am I doing wrong?
Used "set" instead of "export" on windows to provide my API KEy but when I run the app it still says that I have not provided my API key ( i'm subscribed to GPT4). Any one having the same issue ?
For anyone getting the error " CommandNotFoundError: Your shell has not been properly configured to use 'conda activate'." Identify your shell: Check which shell you are using by running the following command: "echo $SHELL" The output will display the path to your current shell. Run the "conda init" command: Based on the output from the previous step, replace in the following command with the name of your shell (e.g., bash, zsh, fish): conda init For example, if your shell is Bash, you would run: "conda init bash"
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?
Note that the author has already changed the way this is run, so be sure to read the README, as even though this video is only 4 days old as of this comment, it's already a little out of date. Also, I tried a slightly more complex project than "snake", and it produced a complete hallucination with gpt-3.5-turbo, since I don't have gpt-4 yet. (I wanted it to bring up a button that played a sound along with a few other specs, and it generated code to find my tax bracket among other things...) Interesting approach, but it's not there yet...
I got the snake game created ... took some fiddling but once I got the right model selected it created it really well. I had to add one two fixes time.sleep() to slow the game down, and import a missing module that didn't get added. but it works amazing!!!
Tried to create the classic 1981 Apple II game Sabotage. Wrote a very detailed prompt based on the Wikipedia description. Then I purchased a Chat-GPT 4.0 API key. It was a total failure! I just got the message: "Due to the complexity of the game, the full implementation of these classes would be quite extensive and beyond the scope of this task."
@@casperd2100 Not i think you can do a lot for 5 $ its credit based you dont need GPT plus ... but yea 8000 token per conversation limit could be a little problem to get good product
if you want to set your API key permanently, open the main py, at the top add 'import openai' without quotes Then before the @ app command put openai.api_key = "your open ai api key" with quotation marks around the api key.
Is there such a thing, with all these Engineering/coding AI assistants, as all the prompts and the generated code being kept and accessible to the chatbot creators, so that they actually harvest all the possible algorithm/code ideas from the users, to use them for themselves as they want ?
If pip barfs on the install, with warning "Could not find a version that satisfies the requirement [module name]"-- check and make sure you have the current python update (3.11 as of the time of this post).
Hi Mathew, Thank you so much for the amazing videos you make. I am running in to the following issue while trying to run it in my windows. --- Processing fstab with mount -a failed. WSL (8) ERROR: CreateProcessEntryCommon:370: getpwuid(0) failed 2 WSL (8) ERROR: CreateProcessEntryCommon:374: getpwuid(0) failed 2 WSL (8) ERROR: CreateProcessEntryCommon:577: execvpe /bin/bash failed 2 WSL (8) ERROR: CreateProcessEntryCommon:586: Create process not expected to return
Matthew, on windows 11 and no requirements.txt was created or found. ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' Want to test this but stuck...
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?
Great video! I am trying this and am stuck at an error at "python -m pip install -r requirments.txt" its coming back with no such file or directory. Thanks!
Top RUclips 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?
in my gpt-engineer folder and it does not have your prompt git:(main) C:\Users\Oscar Gaming\gpt-engineer>conda create --name gpt-eng python 'conda' is not recognized as an internal or external command, operable program or batch file.
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?
It says conda isn't recognized... and before you ask yes I have downloaded, installed and set the environment variables for conda... so what could be the problem?
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
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.
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
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
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!
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.
It is crucial that we provide support for these types of projects to give them the encouragement they need and help them progress further..
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"
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!
Hey,
When I cloned the repo, I am not able to follow along with the steps:
python gpt_engineer/main.py projects/example ─╯
Traceback (most recent call last):
File "/Users/saymukhe/Personal_Code/gpt-engineer/gpt_engineer/main.py", line 10, in
from gpt_engineer.ai import AI
ModuleNotFoundError: No module named 'gpt_engineer'
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?
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!
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.
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.
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.
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 :)
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 ."
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.
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
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.
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
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",
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.
Awesome video. Short, to the point and no detail sacrificed. Cheers
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
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
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
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?
User: Given that LLM like you are able to write software, soon the job of software developer will become outdated. If you were a developer what would you do given that sooner of later all the fields will became more or less automated?
Gpt: As an AI language model, I don't possess personal ambitions or intentions. However, I can provide some guidance on how software developers can adapt to the evolving landscape of automation and AI technologies.
1. Embrace automation: Instead of viewing automation as a threat, embrace it as an opportunity. Learn about emerging tools and frameworks that automate repetitive tasks in software development, such as code generation, testing, and deployment. This allows you to focus on higher-level tasks that require creativity and problem-solving.
2. Cultivate unique skills: As automation takes over routine programming tasks, it becomes essential to develop skills that complement automation. Focus on areas like architecture design, algorithm optimization, data analysis, machine learning, and human-computer interaction. These skills require critical thinking, domain expertise, and creativity that are difficult to automate.
3. Keep learning and adapting: The field of technology is constantly evolving, and staying up to date with the latest advancements is crucial. Continuously learn new programming languages, frameworks, and tools. Explore emerging technologies like blockchain, virtual reality, or Internet of Things (IoT), which are likely to offer new opportunities and challenges.
4. Collaborate and specialize: As software development becomes more automated, collaboration and specialization become increasingly valuable. Engage in cross-disciplinary projects that require collaboration with experts from other domains such as design, business analysis, or data science. Specialize in a specific industry or niche, where domain knowledge is essential and automation is less likely to replace human expertise.
5. Emphasize soft skills: While technical skills are important, soft skills will continue to be highly valuable. Develop skills in communication, problem-solving, leadership, and project management. These skills are essential for effective collaboration, managing teams, and understanding user needs.
6. Adapt to new roles: The role of a software developer may evolve into new areas such as AI ethics, data privacy, or software consultancy. Stay open to exploring new roles and opportunities as the industry changes. Consider pursuing advanced degrees or certifications to enhance your expertise in specific areas.
Remember, automation and AI are tools that can enhance the work of software developers, not completely replace them. By adapting and evolving alongside these technologies, software developers can continue to contribute to the industry and find meaningful roles in the changing landscape.
But yeah you're fucked
Great tutorial, thanks! After a bunch of iteration I was able to get a pretty good working version of breakout going!
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 ?
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
There is no requirements.txt file for pip, anyone else have the issue? It's not on GIT.
Given snake type games have been around for 40 years or so surely an intelligent AI would just output an existing example of the game and then translate it to whatever language you want. How do we know it’s interpreting the prompt and writing the code from scratch as opposed to pulling an example etc.?
App Digital Vending machines incoming 2024. Thanks for keeping the vid short bro.
You got it!
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...
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
i wish this would use a local model instead of the API. but still cool to see. tnx for sharing.
you still need to be the engineer understanding the problem that needs building, to be able to explain it to GPT-engineer or even ChatGPT. Never built a snake game but by just guessing, I can conclude, its relatively simple. Maybe better example would be a Pacman game or Tetris. Those would be more complex. How about a side-scroller game? :P For a toy or a helping tool to give you ideas, yes, its good, but to build a complete software solution with thousands of requirements, I doubt so.
Are you sure that will hold for the future?
@@TheDefender123Plays i am guessing no.
Can't even imagine where this will be in 10 years.
I'm not sure this is a fair look at the functionality -- esp when you say it's the only generator that was able to build the snake game in a single pass. The reason is because it appears the dev optimized this for the snake game -- we can surmise this by the fact it's the example project. I'd love to see you do a couple of other game gens, like maybe flappy bird. Or breakout.
exactly. respect to the dev first and foremost, but there are lots of gaps left when you let the script loose on one of your own ideas. still amazing though.
Yeah as soon as you start tweeking it kind of gets lost. I'd only had 3 runs. First with out tweeks which work but created a over simple game, trying to improve the game with more prompts kind of made broken games when parts where not implemented.
I'd say next step in development would be a review function that can test and fix issues and complete parts that are not written. That would require a more rigid design imo.
Common just found this channel, but if this is the type of content (trying to generate a game that is the example project...), will have to unsubscribe
Love your videos! It is too bad that you need a GPT-4 API Key, still on the waitlist :(
Now, what would be super cool is to drive this from a local LLM. Might be slower, but then you can iterate without breaking the bank.
100%
For some reason after activating conda, when I try "python -m pip install -r requirements.txt" I get this error from the terminal on my macbook: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt'. Do you have a fix for this or what may be occurring? Everything was working fine up until this point.
Is anyone getting this message; ERROR: Could not open requirements file: Errno2 No such file or directory: 'requirements.txt
And am using the Anaconda Prompt(minicomda)
Please help
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.
Hi Mathew, i really would like to figure out how to connect a locally running model, or a model in LM Studio to this. How might I do that?
Very cool, it'll be even more exciting when it doesn't require chat-gpt & can use local & open source modes instead.
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!!
Great, GPT 'engineer' made the equivalent of a high schoolers first python project. Now lets see it do something actually useful.
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?
Absolutely incredible 🤩
Thanks :)
Thank you for the video, I had to modify mine a bit but I did get it working.
when i got as far as 'python -m pip install -r requirements.txt' I had an error that it couldn't find the requirements.txt file. Where do i locate this file? I didn't see it in the repo for GPT-ENG.
Awesome stuff Matt! It's like Christmas everyday.
Haha, I love this. I literally asked if the snake game could be created with SMOL and then this came out.
Bruh you are right on the bleeding edge of all of this and i'm on your coattails thanks!
So well explained thanks very much . Greetings from James J in Limerick city Ireland 🇮🇪
does it generate only python code, or others ? thanks for the video anyway...
I get this when trying to execute python -m pip install -r requirements.txt: ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' What am I doing wrong?
Great Video! Thanks for bringing this project to our attention (and I thought the 16k token news was the pinnacle for this week 😅)
I don't know python but your vids are inspiring - keep it coming :)
Thanks :)
I got an error ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt'
Used "set" instead of "export" on windows to provide my API KEy but when I run the app it still says that I have not provided my API key ( i'm subscribed to GPT4). Any one having the same issue ?
You can probably just edit the env file directly with your API key
Hi, I tried installing conda but no joy, i have tried using CMD and github CMD and am lost as to what im doing wrong
Relevant information. Worked like a charm . Thank u
For anyone getting the error " CommandNotFoundError: Your shell has not been properly configured to use 'conda activate'."
Identify your shell: Check which shell you are using by running the following command:
"echo $SHELL"
The output will display the path to your current shell.
Run the "conda init" command: Based on the output from the previous step, replace in the following command with the name of your shell (e.g., bash, zsh, fish):
conda init
For example, if your shell is Bash, you would run:
"conda init bash"
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?
I thought I was subscribed to more than enough channels on AI, close to a dozen.....but now I just added another 👍
Note that the author has already changed the way this is run, so be sure to read the README, as even though this video is only 4 days old as of this comment, it's already a little out of date. Also, I tried a slightly more complex project than "snake", and it produced a complete hallucination with gpt-3.5-turbo, since I don't have gpt-4 yet. (I wanted it to bring up a button that played a sound along with a few other specs, and it generated code to find my tax bracket among other things...) Interesting approach, but it's not there yet...
Absolutely outstanding! Thanks for sharing such amazing content!
I got the snake game created ... took some fiddling but once I got the right model selected it created it really well. I had to add one two fixes time.sleep() to slow the game down, and import a missing module that didn't get added. but it works amazing!!!
Tried to create the classic 1981 Apple II game Sabotage. Wrote a very detailed prompt based on the Wikipedia description. Then I purchased a Chat-GPT 4.0 API key. It was a total failure! I just got the message: "Due to the complexity of the game, the full implementation of these classes would be quite extensive and beyond the scope of this task."
Awesome Matthew! Doing the research and saving us time. Love it.
Great video Matthew I am definitely going to try this. Thanks!
Great tutorial. I really like this new project. Thanks for making the vid👍
This is really excellent, very concise, and to the point. I presume it also runs on MAC OS environment?
how many tokens the command cost?
Oh good question, I’m not sure, I didn’t check.
@@matthew_bermanYea, I wanted to know how much this costs. Does it use gpt-3.5-turbo?
I just found out it has to use gpt-4 😑 Gonna be so expensive to use
@@casperd2100 Not i think you can do a lot for 5 $ its credit based you dont need GPT plus ... but yea 8000 token per conversation limit could be a little problem to get good product
if you want to set your API key permanently, open the main py, at the top add 'import openai' without quotes
Then before the @ app command put
openai.api_key = "your open ai api key"
with quotation marks around the api key.
Is there such a thing, with all these Engineering/coding AI assistants, as all the prompts and the generated code being kept and accessible to the chatbot creators, so that they actually harvest all the possible algorithm/code ideas from the users, to use them for themselves as they want ?
awesome. we need this kinds of videos
When using windows, how do I get around the OS error no 22 Invalid Argument, where it shows a path to my workspace followed by \
ight:
Is there any similar GPT engine can do the same trick for apple app development for swift program?
If pip barfs on the install, with warning "Could not find a version that satisfies the requirement [module name]"-- check and make sure you have the current python update (3.11 as of the time of this post).
Thanks you.
Hi Mathew,
Thank you so much for the amazing videos you make.
I am running in to the following issue while trying to run it in my windows.
---
Processing fstab with mount -a failed.
WSL (8) ERROR: CreateProcessEntryCommon:370: getpwuid(0) failed 2
WSL (8) ERROR: CreateProcessEntryCommon:374: getpwuid(0) failed 2
WSL (8) ERROR: CreateProcessEntryCommon:577: execvpe /bin/bash failed 2
WSL (8) ERROR: CreateProcessEntryCommon:586: Create process not expected to return
Maybe the author can use embedding sao it can read the code from a vector database. Amazing video btw
And we're just at the early stages. Just imagine what it would be like next year
At which program on mac do you write that?
I am able to clone that, but at which tool to build that?
Matthew, on windows 11 and no requirements.txt was created or found. ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' Want to test this but stuck...
Had the same trouble, turns out it's the "requires.txt" file inside:
gpt-engineer\gpt_engineer.egg-info
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
Matthew teaches so much gpt I'm beginning to wonder if the headshot recording is AI'd as well XD
That is awesome! Thanks for sharing!
Excellent demonstration. Thank you
Great video! I am trying this and am stuck at an error at "python -m pip install -r requirments.txt" its coming back with no such file or directory. Thanks!
Top RUclips 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?
in my gpt-engineer folder and it does not have your prompt git:(main)
C:\Users\Oscar Gaming\gpt-engineer>conda create --name gpt-eng python
'conda' is not recognized as an internal or external command,
operable program or batch file.
Hey I'm using powershell and I'm having problems on the example part. What do I do
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?
Does anyone know what to enter instead of "python main.py example" in the new repository (as of July 14, 2023)
HEEEELLLLPPPPP
It says conda isn't recognized... and before you ask yes I have downloaded, installed and set the environment variables for conda... so what could be the problem?