Could you consider starting your next video with an open-source project, select a few open issues, and demonstrate how this tool can be used to submit a PR that addresses them? It would give a more realistic view of its capabilities!
@@funkahontas Why wouldn't it be able to use te Github API to make a PR? It's no different to performing any other activity on a local machine, it just needs to do it with API calls.
It should be amazing if it could use another LLM like Mixtral or Deepseek Coder. Or a combination of 2 models, maybe Deepseek or CodeLllama for coding and Mistral or other general Llama for comprehension and planning.
@@matthew_berman great news. it is open source. you don't need to add that functionality. someone else will do that. probably these are the last few years where mankind programs open source programs. soon open source AI will program open source programs. 😀
In theory this should not be that difficult for GPT-Pilot to support especially when you use something like LM Studio and create a local API server connected to open source models on HF
Would be a nice project to use GPT copilot to add this feature to itself. But I fear that is beyond the abilities of this tool at the moment, because it is not as good at creating stuff that it has not seen a million times before. Login frontend and database in the backend is in almost every full stack app and there are lots of tutorials on it. Completly new ideas might be harder for the tool, especially if you as a humen do not know exactly what you want it too build. But I might be wrong let me know if someone tried this.
Great vid! How about a GPT Pilot vs. GPT Engineer comparison next? The latter has 2x the stars on github, higher commit frequency, active community, etc. You did a comparison a while back. I'm curious to how things stack up now.
This is a great video demonstration of GPT-Pilot and Matthew does it so effortlessly ... which is a problem because it raises expectations that cannot be met. The truth is that this is an alpha version and is buggy, and depending on your OS and the programming language you use, it may work or break immediately or not work at all. The most stable is the combination of Mac OS and Python. Obviously because the authors code on Mac and use Python. What might crash more often than not is the VSCode extension on Windows. And it's not that good with frameworks. So keep your expectations low and don't expect GPT-Pilot to write the perfect fullstack application to conquer the world - it's not capable of that - yet ;-)
Im a full stack dev, while the code might work, the depreciation warnings and request to manually add the bootstrap files (rather than import the npm package) indicates to me that the code quality might be very poor. This is incredibly cool but I wouldn’t rely on it for any serious projects just yet
Eventually it probably won’t matter how “well written” the code is, once the AI is the only one who has to read it. In the meantime though, still an important factor in practice.
@@ai-lucas So you are saying performance does not matter? lol There are a ton of things in this where it's writing bad code... If people are saying "Yeah its great" tells me they are not a true engineer.
@@ai-lucas I have to disagree, as @helix8847 mentioned, bad code can mean slow performance leading to needless extra server costs assuming it doesn’t entirely crash when scaled. But using deprecated libraries/code can easily introduce unpatched vulnerabilities into your system which I’d argue is even more detrimental
Hi Matt! I really love watching your videos and coding along. In the last year, thanks to you and chat-gpt (and the others) I have really levelled up as a programmer (have watched other channels and studied of course!) but your super enthusiastic "Today!..." really really makes me happy and excited. Wanted to thank you for your amazing content and ask you to PLEASE PLEASE create an implementation video of GPT-Pilot with an open source model that runs well on a basic M1 Macbook Air (the stuff most of your viewers are actually using!) so that we can get a jumpstart in our dev journey! Thanks again in case do do it, if not will eagerly await your next piece!
Please make a general tutorial on using tools like this, GPT-Engineer, aider, smol developer, mentat, using the OpenAI API system, but using local LLMs instead. And what are the best Local LLMs (making a leaderboard) to work with them, depending on things like precision, stability and context window size.
It's super good when you are starting a brand new project, but I am still trying to find a good one to deal with existing projects, I have around 40 small services under an NX environment and it would be amazing to have a code assistant to understand this workspace and be able to make modifications to it. If you guys know any good for it that I am missing, please shout it out :)
This is the limitation I'm hitting at the moment as well. Until you can have a higher-level "management agent" with a deeper meta-understanding of the project's moving parts than is currently possible with LLMs, we're stuck. I have NO DOUBT that it's coming very soon, though, given the pace of innovation we're seeing right now.
@@FrankEBaileyAbsolutely, no complaints from my end! It's pretty awesome to be part of this tech revolution, especially when I think back to the days of coding on my ancient Solaris. Right now, I'm using Copilot with VS Code and JetBrains AI with IntelliJ, and honestly, IntelliJ is giving me better vibes. I think it's because IntelliJ has this tight grip on the whole project, so it prompts the LLM more effectively. This way, I'm spending less time trying to figure out the perfect prompts and more time actually coding."
@@FrankEBailey @ishtyleretienne36 Have you played around with Aider? It takes a snapshot of the project library for context and can create files, write code etc
Aider But it’s so much cheaper and faster to just rewrite everything from scratch. It was always easier, but now… Just don’t tell your boss, he won’t understand.
$20 is for the chat, the API will cost you a certain amount per 1000 tokens, depending on the model. The most advanced and expensive one (GPT-4 Turbo) costs 0.01 per 1000 input tokens and 0.03 per 1000 output tokens.
I gotta say... I've been a product manager for the last 13 years or so. I have no idea how to code, but I understand the concepts and can write somewhat technical requirements. Even before GPT Pilot, I've been able to build entire apps using AI (first GPT-4, then Cursor). This looks super interesting...
This is great tech and will only get better, I have my own project that does something similar. I like some approaches used to interact with the human. It would be great to see in real time how many tokens you are using, and show usage in $ dollar amounts, and for debugging to see the raw api requests.
this system has a ways to go for anything with even a mild amount of complexity outside of what gpt normally programs with (think slightly more niche libraries or frameworks with less documentation ) hell even just handling git control my github pilot tended to repeat the same tasks of confirmation that were unnecessary and managed to fill up my api limit from 0 to $3.77 with in an hour. it is a very cool idea and will work for most mainstream small tasks but then again I can do that just as easily for free. It will be better in the future and is immensely great considering it is open source.
It depends. For a decently large application, which might takes a few hours to build, you can expect $10-15 in API charges from OpenAI. Results may vary!
@@matthew_berman That's a bit steep! One would have to think twice before signing up for OpenAI's API. Alternatives will come out that will probably have flat monthly rates instead.
$10-$15 for an app is far from expensive, but I get it, things will get even cheaper. Great time and opportunity to launch apps for a fraction of what a team of developers would cost.
Hey Matt, my name is David McCarty. I absolutely love your channel. You explain all of this A.I. insanity in a simple and easy to understand way. I have transitioned to watching your channel over all other channels for this reason. I am currently building an app with an industry leading company out of Atlanta however I won't be launching for several weeks. I have been approached by several people wanting me to build for them, but I am too busy with my own app to build theirs. This is where your video on GPT Pilot comes in. I have a few questions about being able to access key components of the app. Are you willing to help explain to me further how this program works like you better explain things in your videos? If interested, is this a good way means of communication, or would it be best to do so through email? Regardless of your decision, I love what you are doing and will continue to learn from you and enjoy watching your genius.
Whats unreal is how much $$$ this uses for very simple tasks. Hopefully they implement the ability to use other models. Once the project reaches any sort of complexity this quickly becomes obsolete. Wasting more time and money on api requests (and looping the same few 'problems' over and over) than manually crafting prompts elsewhere. GPT Plus and other local LLMs produces similar results, albeit slower, without the 'automatic' 'planning and structure'. The only use case (besides newbie programmer nuance )would be to kickstart a project. Once the project reaches any sort of complexity, manual intervention is required. I followed the prompts chosen in the video (omitting a few changes due to os and browser differences) came out to $6 in api charges. For arguably worse results than using a gh react template lol.
lol the first CRAY supercomputer cost $38 million (in todays money), it had 8mb of memory, 300mb of storage and a measly 160 mflops, it weighed 5.5 tons and required 115 kW to run. My computer is tens of thousands of times faster than this supercomputer, it runs on about 200 watts and cost < $1500, same thing will happen with neural nets.@@afterglowdev7905 Also local neural nets cost practically $0 to run.
@@afterglowdev7905 What you see is about a day of dev time, or $1200. That adds up over time. Ultimately you can get your own AI desktop for $20k and run the open source models and then fire all programmers and replace them with product managers making $0.25/hr. Since this is open source you can just modify a few lines of code to talk to your local machine AI. Combine that with these multimodal LLMs that are almost on par with gPT4, and memGPT and you can easily handle complex cases.
Absolutely astounding! This is undeniably the most remarkable tool you've recommended so far! Your dedication and expertise shine brightly in your work. Keep up the exceptional work, and let's continue to push the boundaries of innovation! 🚀
@@Danuxsy a good full stack dev would re-create that (and much more) using a framework and a single command that would generate code that he is familiar with and he can start building more advanced components than the generic auth/login stuff... On his/her own in a few seconds. So no, not only 2-3$ for this that a single dev can generate using the right tool in a few seconds is not 100Kx time cheaper (not even faster!). This is no where near what you want to replace a full stack dev team working on a real product. YET... Soon however (imho 2-3 yrs) it will be.
in the 2030s these systems will probably be able to generate source code for entire games like World of Warcraft, milions of lines of code in 24 hours or smh like that. @@sinitarium
Christ, that was impressive! I am an old school emacs lamer, but this is seriously making me consider putting vscode on one of my Linux systems and trying this out.
Unfortunately, after two days and 15 $ worth of API calls, I must say that I have been unable to get to build even the simplest web app, trying different technologies. For example, it constantly gets confused with the directory structure (which it builds as mapp\myapp\myapp) and tries to access files on the wrong level of the myapp-hierarchy. Or it just generates a file and then tells you to add the code. At this stage, it is a tool for experienced developers to do some routine stuff for them, I guess.
I looked over the code and docs and that's not currently supported. That said, if you can reasonably spin up several agents, you could definitely modify the code to point to and interact with local models instead
I start it up, it asks me to sign up. I enter my email, then halfway through entering my password it navigates away. I don't know who is doing the navigating. VS Code stuff 🤔
Nothing to do with your content but I thought it was funny..,. When looking at my YT page suggestions I saw your video thumbnail and said out loud to myself... "Why is Trent Horn on my page but doing AI stuff...?" (I follow both of you) Great video by the way lol
@@zvone187 Hi zvone, I'm stuck at the very beginning, when I clicked create new app I got a red text line saying "The system cannot find the path specified" although I already had the path to C:\Users\userName\gpt-pilot\workspace , what am i missing here, please?
I feel like in the future, folks will forget how me you and other developers use to do this manually. We might get a future where code is generated and people don't quite understand it since they were so use to the AI generating it. I mean for now there are folks like us, but what about the future? Kind of an interesting thought anyway.
Already like 30% of college age students don't know how to move a file from one folder to another because they have only ever used phones with apps. There are even people who don't know how to use a can opener...
So I just started an Angular project, should be a landing page ... but never came because of mistakes. And the whole thing has already cost 3 EUR in about 7 minutes. So the hourly rate is already impressive: D and without knowing whether something usable will come out in the end ...
Off the topic, what was the name of the app that makes this effect you are using in the start of the video, had it before on iPhone but cannot remember? BTW, great video! :D
Unfortunately not at this point. We're focused on creating a base for the AI to iterate on a codebase. Once that is working, we believe it will be relatively trivial to add it to work on an existing codebase by analysing a codebase and asking questions about the code.
I did get it to work, but it wasn't nearly as impressive when trying to use it with Python + Flask. Every step of the way I had to make manual modifications to get it to run. By the end, it had forgotten most of the user stories it had written. Like a lot of alpha software, it probably needs a bit more love before it's ready for wider use.
Do youhave some stats of the gpt4 api consumption of the app you built? It would be interesting to see how many calls it did and how many tokens I/O it used
@@FranPetraroli it makes sense, the amount of api calls and the size of the inputs and outputs should be quite big. I thought that was the case, not i guess is confirmed
I've installed it from Github and tried the command line version and I think it's underwhelming, if not saying pretty useless -- constant struggling with the project structure, compiler errors etc. Designing the application yourself and just letting the LLM provide you with smaller, clearly defined functions is still much better than this.
Well actually you can do this on gpt-3 too this is just making the process easier and faster because you’re not copy pasting codes around etc. But I am curious if gpt-4 can make ui visuals. Because all it does creating the interface but it doesn’t add visual props etc… If you know anything please enlighten us ✌🏼
Hello, Would you be able to crate a video that provides the information about "Free" api keys from openai and there usage? The current documentation is kind of confusing and convoluted and thus don't know if I have an available "Free" use of the keys or not. I ask this because I get errors when trying to use the keys, even though I've never used them before... Thanks,
These things are great, but anyone with enough experience working on real life systems will tell you, using this will become a maintenance nightmare - no real world application is build and forget, greenfield development is fun because it’s ‘easy’, the real work comes from extending, debugging, optimising and like and that all requires a good and understood architecture that you will totally fail to get using a GPT The real goldmine in these early days of AI hype will be the work required to understand all the code that these dice rollers have generated
i disagree. ai is a tool unlike anything we've seen before! and at the rate it's increasing it will certainly get better. this particular tool will learn literally from the feedback you just provided and know what needs to do. it's hard to believe we're alive to witness a technology like this
with RAG and customGPTs that are covering whole documentation of particular technology on top of this planner, I do not see any limits, especially in python which chatgpt nails in 99% cases
Cool demo but it does not work in practice. Its crashed on me 5 times and i have a modern machine. Even reinstalled vs code to test again and same thing. I think this is just too early. But definitely cooltho!
It keeps doing two things when i run it in my Code VS. It keeps telling me that I don't have an ALS debugger and asks if I want to install one. I have installed a total of 12 debuggers and it still does not work. Also, it keeps defaulting to version of ChatGPT-4- 1106-Preview which is foreign to me. Useless and frustrating at this point.
Could you add some more info: - what LLM must be used and if GPT-4, how much was the cost of this? - can GPT Pilot access latest library docs, so does it do some RAG? Thanks
You can use any LLM with GPT Pilot but GPT-4 will work the best by far. Currently, it cannot access any external documentation but that will be implemented very soon.
Is there any way to not use openAI but a local llm, mistral or so ? This would be a true gamechanger. With openAI it’s getting too expensive to even figure out how this should be done properly ..
Does it ever hang for you? I've had it get stuck on Waiting while coding, spinning wheel forever, and a couple other critical issues Not sure what to do when that happens. If I go back, it will hang at the same spot every time.
![โอ้เธอช่าง... - บี้เดอะสกา ft. ต๋อง เทวัญ (Prod. by ป๋าเพชร) [Official MV]](http://i.ytimg.com/vi/zbWhiLFmVL0/mqdefault.jpg)
![[Part 1] ทารกน้อยถูกทิ้ง👶 หญิงชราเก็บขวดขายเลี้ยงดูเขาจนโต #โลกการละคร](http://i.ytimg.com/vi/nPTDvFCExks/mqdefault.jpg)
This is one of the most impressive AI coding tools I've seen so far. Let me know what you think!
wow sick! This is the future of development! I am using github copilot, but copilot is a joke in comparison to this!
I don’t suppose it can be used for MacOS development 😀 ? That’s one thing I keep hoping to see.
As always, thank you for sharing! 🔥
Looks suspicious. Sign-up doesn't work. Also discord link is not working either.
Edit: it was a issue that got fixed the next day.
How much did this app cost in gpt4 tokens?
Could you consider starting your next video with an open-source project, select a few open issues, and demonstrate how this tool can be used to submit a PR that addresses them? It would give a more realistic view of its capabilities!
It cannot do that. Delusional
@@funkahontas Why wouldn't it be able to use te Github API to make a PR? It's no different to performing any other activity on a local machine, it just needs to do it with API calls.
AI isn't capable of that, and really neither is Matthew Berman.
I'm already using AI to do this 😅 I don't know why people think it can't
Any PR links to share?@@jnevercast
It should be amazing if it could use another LLM like Mixtral or Deepseek Coder. Or a combination of 2 models, maybe Deepseek or CodeLllama for coding and Mistral or other general Llama for comprehension and planning.
especially with custom system prompts on mixtral/codewizard 34b
GPT Pilot is open source, so I'm sure you can add that functionality ;)
@@matthew_berman great news. it is open source. you don't need to add that functionality. someone else will do that.
probably these are the last few years where mankind programs open source programs. soon open source AI will program open source programs. 😀
In theory this should not be that difficult for GPT-Pilot to support especially when you use something like LM Studio and create a local API server connected to open source models on HF
Would be a nice project to use GPT copilot to add this feature to itself. But I fear that is beyond the abilities of this tool at the moment, because it is not as good at creating stuff that it has not seen a million times before. Login frontend and database in the backend is in almost every full stack app and there are lots of tutorials on it. Completly new ideas might be harder for the tool, especially if you as a humen do not know exactly what you want it too build. But I might be wrong let me know if someone tried this.
Great vid! How about a GPT Pilot vs. GPT Engineer comparison next? The latter has 2x the stars on github, higher commit frequency, active community, etc. You did a comparison a while back. I'm curious to how things stack up now.
I think Aider outperforms them
@@DJPapzin Aiders approach is mostly the same as the GTP Engineer improve approach
This is a great video demonstration of GPT-Pilot and Matthew does it so effortlessly ... which is a problem because it raises expectations that cannot be met. The truth is that this is an alpha version and is buggy, and depending on your OS and the programming language you use, it may work or break immediately or not work at all. The most stable is the combination of Mac OS and Python. Obviously because the authors code on Mac and use Python. What might crash more often than not is the VSCode extension on Windows. And it's not that good with frameworks. So keep your expectations low and don't expect GPT-Pilot to write the perfect fullstack application to conquer the world - it's not capable of that - yet ;-)
Im a full stack dev, while the code might work, the depreciation warnings and request to manually add the bootstrap files (rather than import the npm package) indicates to me that the code quality might be very poor.
This is incredibly cool but I wouldn’t rely on it for any serious projects just yet
Eventually it probably won’t matter how “well written” the code is, once the AI is the only one who has to read it.
In the meantime though, still an important factor in practice.
@@ai-lucas So you are saying performance does not matter? lol
There are a ton of things in this where it's writing bad code... If people are saying "Yeah its great" tells me they are not a true engineer.
@@ai-lucas I have to disagree, as @helix8847 mentioned, bad code can mean slow performance leading to needless extra server costs assuming it doesn’t entirely crash when scaled.
But using deprecated libraries/code can easily introduce unpatched vulnerabilities into your system which I’d argue is even more detrimental
@@ai-lucas I respectfully disagree. Poor code and out of date libraries result in security flaws. That's how you get hacked.
@@helix8847 I said eventually.
We got a ways to go, but this is definitely the right direction. The interactivity and the ai prompting the user is impressive.
Hi Matt! I really love watching your videos and coding along. In the last year, thanks to you and chat-gpt (and the others) I have really levelled up as a programmer (have watched other channels and studied of course!) but your super enthusiastic "Today!..." really really makes me happy and excited.
Wanted to thank you for your amazing content and ask you to PLEASE PLEASE create an implementation video of GPT-Pilot with an open source model that runs well on a basic M1 Macbook Air (the stuff most of your viewers are actually using!) so that we can get a jumpstart in our dev journey!
Thanks again in case do do it, if not will eagerly await your next piece!
As always very thorough presentation, thank you.
May I ask if you can share how much did it cost on openAi usage for creating this app?
Please make a general tutorial on using tools like this, GPT-Engineer, aider, smol developer, mentat, using the OpenAI API system, but using local LLMs instead. And what are the best Local LLMs (making a leaderboard) to work with them, depending on things like precision, stability and context window size.
Really nice. My problem is - as usual - no info about the cost of OpenAI for such tasks.
It's super good when you are starting a brand new project, but I am still trying to find a good one to deal with existing projects, I have around 40 small services under an NX environment and it would be amazing to have a code assistant to understand this workspace and be able to make modifications to it. If you guys know any good for it that I am missing, please shout it out :)
GitHub copilot's new functionality is supposed to help with this. Have you given that a shot?
This is the limitation I'm hitting at the moment as well. Until you can have a higher-level "management agent" with a deeper meta-understanding of the project's moving parts than is currently possible with LLMs, we're stuck. I have NO DOUBT that it's coming very soon, though, given the pace of innovation we're seeing right now.
@@FrankEBaileyAbsolutely, no complaints from my end! It's pretty awesome to be part of this tech revolution, especially when I think back to the days of coding on my ancient Solaris. Right now, I'm using Copilot with VS Code and JetBrains AI with IntelliJ, and honestly, IntelliJ is giving me better vibes. I think it's because IntelliJ has this tight grip on the whole project, so it prompts the LLM more effectively. This way, I'm spending less time trying to figure out the perfect prompts and more time actually coding."
@@FrankEBailey @ishtyleretienne36 Have you played around with Aider? It takes a snapshot of the project library for context and can create files, write code etc
Aider
But it’s so much cheaper and faster to just rewrite everything from scratch. It was always easier, but now…
Just don’t tell your boss, he won’t understand.
Hey can you share how much did it cost to make that app ?
$20 is for the chat, the API will cost you a certain amount per 1000 tokens, depending on the model. The most advanced and expensive one (GPT-4 Turbo) costs 0.01 per 1000 input tokens and 0.03 per 1000 output tokens.
@@swdev245 I think he's asking how much did it cost on token calls to gpt4
I'd love to hear from Matt how much the GPT 4 cost was on this project. Seems like that cost could add up very quickly
Please tell us Matt!
@@Fuzzhead93 probably a few dollars at most.
I gotta say... I've been a product manager for the last 13 years or so. I have no idea how to code, but I understand the concepts and can write somewhat technical requirements. Even before GPT Pilot, I've been able to build entire apps using AI (first GPT-4, then Cursor). This looks super interesting...
what did you build? any ;ove example?
Suuuure you are and did lmao
This is great tech and will only get better, I have my own project that does something similar. I like some approaches used to interact with the human. It would be great to see in real time how many tokens you are using, and show usage in $ dollar amounts, and for debugging to see the raw api requests.
These are great ideas! Btw, if you have a project you've created, I'd love to check it out. Can you share it?
Same here. I really wonder how many of us are building something similar. It's hard to estimate.
this system has a ways to go for anything with even a mild amount of complexity outside of what gpt normally programs with (think slightly more niche libraries or frameworks with less documentation ) hell even just handling git control my github pilot tended to repeat the same tasks of confirmation that were unnecessary and managed to fill up my api limit from 0 to $3.77 with in an hour. it is a very cool idea and will work for most mainstream small tasks but then again I can do that just as easily for free. It will be better in the future and is immensely great considering it is open source.
Would have been interesting to see how it handles postgres
There's actually a postgres branch. Not sure what state it's in.
Can you test a similar project (if there is any) that works with open-source AIs? unless there is a way of using this but with opensource AIs
Yes, you can use any of the open source models with OpenRouter - just keep in mind that it works by far the best with GPT-4.
incredible !!!
Well done Matt
Imagine GPT Pilot with Open Chat 3.5 or Mistral!
That would be very cool
@@HackerSkid5G you probably can do this, as the gpt4all api is 1:1 the same as from openai.
Curious how much it cost (meaning GPT4 API costs)
It depends. For a decently large application, which might takes a few hours to build, you can expect $10-15 in API charges from OpenAI. Results may vary!
@@matthew_berman That's a bit steep! One would have to think twice before signing up for OpenAI's API. Alternatives will come out that will probably have flat monthly rates instead.
@@Gabeyreit's steep if you're experimenting with the tool, not that much if you're actually building up and saving many working hours
$10-$15 for an app is far from expensive, but I get it, things will get even cheaper. Great time and opportunity to launch apps for a fraction of what a team of developers would cost.
If you used a human that price would be in the hundreds of dollars range instead lol@@Gabeyre
Hey Matt, my name is David McCarty. I absolutely love your channel. You explain all of this A.I. insanity in a simple and easy to understand way. I have transitioned to watching your channel over all other channels for this reason. I am currently building an app with an industry leading company out of Atlanta however I won't be launching for several weeks. I have been approached by several people wanting me to build for them, but I am too busy with my own app to build theirs. This is where your video on GPT Pilot comes in. I have a few questions about being able to access key components of the app. Are you willing to help explain to me further how this program works like you better explain things in your videos? If interested, is this a good way means of communication, or would it be best to do so through email? Regardless of your decision, I love what you are doing and will continue to learn from you and enjoy watching your genius.
I am a developer and it's unreal what it can do. Just imagine in a couple of years.
Whats unreal is how much $$$ this uses for very simple tasks. Hopefully they implement the ability to use other models. Once the project reaches any sort of complexity this quickly becomes obsolete. Wasting more time and money on api requests (and looping the same few 'problems' over and over) than manually crafting prompts elsewhere. GPT Plus and other local LLMs produces similar results, albeit slower, without the 'automatic' 'planning and structure'. The only use case (besides newbie programmer nuance )would be to kickstart a project. Once the project reaches any sort of complexity, manual intervention is required. I followed the prompts chosen in the video (omitting a few changes due to os and browser differences) came out to $6 in api charges. For arguably worse results than using a gh react template lol.
thats reassuring.....I mean I thought AI had made all coding/written language related professions obsolete@@afterglowdev7905
lol the first CRAY supercomputer cost $38 million (in todays money), it had 8mb of memory, 300mb of storage and a measly 160 mflops, it weighed 5.5 tons and required 115 kW to run. My computer is tens of thousands of times faster than this supercomputer, it runs on about 200 watts and cost < $1500, same thing will happen with neural nets.@@afterglowdev7905 Also local neural nets cost practically $0 to run.
@@afterglowdev7905 What you see is about a day of dev time, or $1200. That adds up over time. Ultimately you can get your own AI desktop for $20k and run the open source models and then fire all programmers and replace them with product managers making $0.25/hr. Since this is open source you can just modify a few lines of code to talk to your local machine AI. Combine that with these multimodal LLMs that are almost on par with gPT4, and memGPT and you can easily handle complex cases.
In a couple of years, if you haven't stepped up your game, jobs are gone!
Absolutely astounding! This is undeniably the most remarkable tool you've recommended so far! Your dedication and expertise shine brightly in your work. Keep up the exceptional work, and let's continue to push the boundaries of innovation! 🚀
This looks great. Downloading now and trying it out.
Impressive. How much in openai API cost ?
$2-$3 seems to be the word on the street
hundreds of times less than a dev team doing the same thing.
@@Danuxsy a good full stack dev would re-create that (and much more) using a framework and a single command that would generate code that he is familiar with and he can start building more advanced components than the generic auth/login stuff... On his/her own in a few seconds.
So no, not only 2-3$ for this that a single dev can generate using the right tool in a few seconds is not 100Kx time cheaper (not even faster!). This is no where near what you want to replace a full stack dev team working on a real product. YET...
Soon however (imho 2-3 yrs) it will be.
in the 2030s these systems will probably be able to generate source code for entire games like World of Warcraft, milions of lines of code in 24 hours or smh like that. @@sinitarium
how does it help in existing app?
I’ve tested it and I have to say that there are a lot of bugs to fix. But it’s still impressive.
It’s especially good at building with Express JS.
Christ, that was impressive! I am an old school emacs lamer, but this is seriously making me consider putting vscode on one of my Linux systems and trying this out.
it crashes like hell. 0 stars. Uninstalling gpt-pilot.
Apparently it doesn't work for other people too.
Excellent demo. Love your work.
Unfortunately, after two days and 15 $ worth of API calls, I must say that I have been unable to get to build even the simplest web app, trying different technologies. For example, it constantly gets confused with the directory structure (which it builds as mapp\myapp\myapp) and tries to access files on the wrong level of the myapp-hierarchy. Or it just generates a file and then tells you to add the code. At this stage, it is a tool for experienced developers to do some routine stuff for them, I guess.
Same, but after months. It makes me feel like this is all one big pyramid scheme to get us spending money on open ai
this is exactly what I was doing with AI already, but the workflow will be so much more nice with this.
Hallo :-) D frog isch, was kost so a app wenn für die token zahlsch. Und schriebt dir des ding oh python apps etc? Oder nur js?
Good coding solution. How much did GPT API bill you for this project?
The future solution is obviously to run local neural nets, those are essentially free to run 24/7
Man that was amazing!, and you are a monster when it comes to coding too 🙌👏👍
That is the fiture of coding right there. Have a great christmas 🎄🌟🎁
Hey Matthew, which screen recording software do you use to make these videos?
I wonder if I can use an opensource model with this instead of gpt4?
Same
I looked over the code and docs and that's not currently supported. That said, if you can reasonably spin up several agents, you could definitely modify the code to point to and interact with local models instead
I start it up, it asks me to sign up. I enter my email, then halfway through entering my password it navigates away. I don't know who is doing the navigating. VS Code stuff 🤔
Aider is awesome but this... Looks even more interesting.
Nothing to do with your content but I thought it was funny..,. When looking at my YT page suggestions I saw your video thumbnail and said out loud to myself... "Why is Trent Horn on my page but doing AI stuff...?" (I follow both of you) Great video by the way lol
has anyone actually gotten this to work? everyone testing seems to be stuck at making an account (me included)
Hi, founder of GPT Pilot here. Sorry about this - we're working on it. Can you tell me what happens that you can't sign up?
Hey @michaelpiper8198, we fixed the issue so it should be working now.
@@zvone187
Great work man 🎉
@@zvone187 Hi zvone, I'm stuck at the very beginning, when I clicked create new app I got a red text line saying "The system cannot find the path specified" although I already had the path to C:\Users\userName\gpt-pilot\workspace , what am i missing here, please?
@@userrjlyj5760g same here - C:\Users\userName\gpt-pilot\workspace, tried C:\Users\userName\gpt-pilot as well. No dice.
This is amazing. Where will we be in a year?
I feel like in the future, folks will forget how me you and other developers use to do this manually. We might get a future where code is generated and people don't quite understand it since they were so use to the AI generating it. I mean for now there are folks like us, but what about the future? Kind of an interesting thought anyway.
Already like 30% of college age students don't know how to move a file from one folder to another because they have only ever used phones with apps. There are even people who don't know how to use a can opener...
sign up works for me now. I will give it a shot and see how it goes.👍
stuck in project description and shows waiting. will revoke api key and try again another time.
Nice so essentially alter the API calls to use Huggingface and add say.. mixtral and code llama.. Nice useful program.
So I just started an Angular project, should be a landing page ... but never came because of mistakes. And the whole thing has already cost 3 EUR in about 7 minutes. So the hourly rate is already impressive: D and without knowing whether something usable will come out in the end ...
Off the topic, what was the name of the app that makes this effect you are using in the start of the video, had it before on iPhone but cannot remember? BTW, great video! :D
Great video and amazing content!
Can it work with existing code base/project?
Aways enjoy your content. Thanks and Happy xmas. Terence Nelson New Zealand.
Can you start using it mid-project? Or do you have to start the project with it, for it to work?
Unfortunately not at this point. We're focused on creating a base for the AI to iterate on a codebase. Once that is working, we believe it will be relatively trivial to add it to work on an existing codebase by analysing a codebase and asking questions about the code.
This is freaking amazing!
What HOW MANY LINES of code were written in total? Over 1000? And how many files were made?
I did get it to work, but it wasn't nearly as impressive when trying to use it with Python + Flask. Every step of the way I had to make manual modifications to get it to run. By the end, it had forgotten most of the user stories it had written. Like a lot of alpha software, it probably needs a bit more love before it's ready for wider use.
I think 2024 will be the year we begin to see more robust AI systems
stuck with the path issue, "the system cannot find the specified path". Does anyone else have the same issue?
Yep, same here. Still troubleshooting...
same here...
The path shown is not the path that's actually in the .env file. I had to edit the .env file directly with the correct path.
This is beautiful!!
Do youhave some stats of the gpt4 api consumption of the app you built? It would be interesting to see how many calls it did and how many tokens I/O it used
It 😢 me 6$ to replicate this tutorial, so it's not cheap
@@FranPetraroli it makes sense, the amount of api calls and the size of the inputs and outputs should be quite big. I thought that was the case, not i guess is confirmed
I've installed it from Github and tried the command line version and I think it's underwhelming, if not saying pretty useless -- constant struggling with the project structure, compiler errors etc.
Designing the application yourself and just letting the LLM provide you with smaller, clearly defined functions is still much better than this.
Well actually you can do this on gpt-3 too this is just making the process easier and faster because you’re not copy pasting codes around etc. But I am curious if gpt-4 can make ui visuals. Because all it does creating the interface but it doesn’t add visual props etc… If you know anything please enlighten us ✌🏼
i wonder, if GPT Pilot can write project in any language or only specific ? for example can it create instagram clone in c++ or even in asm?
awesome but why is the Save button disabled for this video?
Wow, very impressive!
What server OS were you using or conneting to?
GPT Pilot requires me to enable/start every system component manually, e.g. mongod etc.
Does it work well though?
If I watched your youtube on making full-stack apps(Free "devin' alternative) and got Pythagora, do I still need to get an account with GPT Pilot?
Please show how to use a local LLM with GPT pilot. Otherwise I see how this can quickly get very expensive.
Hello,
Would you be able to crate a video that provides the information about "Free" api keys from openai and there usage?
The current documentation is kind of confusing and convoluted and thus don't know if I have an available "Free" use of the keys or not.
I ask this because I get errors when trying to use the keys, even though I've never used them before...
Thanks,
This is impressive
Older Mark Norman!!! You've done it again ☺️
probably mark norman is older i would say.
Simply amazing!
Needed new tutorial gpt pilot with local llm. I don't can find the settings of gpt -pilot in vs code plugin((((
It's Pythagora now
These things are great, but anyone with enough experience working on real life systems will tell you, using this will become a maintenance nightmare - no real world application is build and forget, greenfield development is fun because it’s ‘easy’, the real work comes from extending, debugging, optimising and like and that all requires a good and understood architecture that you will totally fail to get using a GPT
The real goldmine in these early days of AI hype will be the work required to understand all the code that these dice rollers have generated
i disagree. ai is a tool unlike anything we've seen before! and at the rate it's increasing it will certainly get better. this particular tool will learn literally from the feedback you just provided and know what needs to do. it's hard to believe we're alive to witness a technology like this
with RAG and customGPTs that are covering whole documentation of particular technology on top of this planner, I do not see any limits, especially in python which chatgpt nails in 99% cases
I am a backend lead working in a startup . This is helpful for my upcoming saas proof of concept. 😂😂.
step in interesting direction. Not there yet, but definitely worth observing.
This is a standard app. It does not mean much. You could find the code online and copy it.
You needed to know quite a lot for it to work. How long would you guess until you’re not needed to know technical things?
This feels illegal to know 😂--SUBBED!
welcome!
@@matthew_berman hey thanks for the warm welcome !🙏
Can it write in Clarity/Svelt?
@@matthew_berman Whats up man?
More Christmas gifts from Matthew..... Can't wait to unwrap this gift and play with it 😊.
Cant create or login into account in vscode plugin... nothing happens when click sign up(((( really alpha
Did you find a fix yet?
@@Mitchg-k6f not yet
same for me, any fixes?
I Found a fix for that! Just use longer passwords. It gives no error if you use weak password but do not let you register.
Hey @allvisualmedia7575, we fixed the issue so it should be working now.
Okay, but how much did the GPT-4 API access cost?
Can it look at a figma file or image, written mockup image or URL and pick up on look and feel of the app that it's creating?
Cool demo but it does not work in practice. Its crashed on me 5 times and i have a modern machine. Even reinstalled vs code to test again and same thing. I think this is just too early. But definitely cooltho!
More Amazing progress...
Cool video. Thanks
I've installed GPT Pilot VS code extension and it asks me for python file giving no explanations, wtf?
Same problem with the path when trying to create a new project
It keeps doing two things when i run it in my Code VS. It keeps telling me that I don't have an ALS debugger and asks if I want to install one. I have installed a total of 12 debuggers and it still does not work. Also, it keeps defaulting to version of ChatGPT-4- 1106-Preview which is foreign to me. Useless and frustrating at this point.
is it limited to gpt4 or can it cooperate with any llm that is good in coding?
It is not limited to GPT-4, and it can be used with any LLM, but GPT-4 will work the best by far.
Is GPT pilot still available? Can't find it in VSC extensions
It's Pythagora now
I hope this can someday join a project mid dev too.
It won't let me sign up... Discord link is broken, can't find the server. What do I do?
same here
same here too
Hi, founder of GPT Pilot here. Sorry about this - we're working on it. Can you tell me what happens that you can't sign up?
The same problem :(
@@zvone187 seems to be working now. but earlier it just showed login/sign up pages and when you entered data into the forms they were non responsive
Could you add some more info:
- what LLM must be used and if GPT-4, how much was the cost of this?
- can GPT Pilot access latest library docs, so does it do some RAG?
Thanks
You can use any LLM with GPT Pilot but GPT-4 will work the best by far. Currently, it cannot access any external documentation but that will be implemented very soon.
Hi, Cannot download the GPT-Pilot extension for chrome. A click on the link says the site cannot be reached.
Love it!
Very impressive. I wish to see it creating a snake game. How much did it cost you to pay OpenUI?
I can't find gpt-pilot visual studio code, is it still available? I see many in the comments saying it was very buggy.
Is there any way to not use openAI but a local llm, mistral or so ? This would be a true gamechanger. With openAI it’s getting too expensive to even figure out how this should be done properly ..
How much does it cost in terms of GPT's tokens & USD?
Does it ever hang for you? I've had it get stuck on Waiting while coding, spinning wheel forever, and a couple other critical issues Not sure what to do when that happens. If I go back, it will hang at the same spot every time.
how are warnings just okay for brand new code with no significant dependency constraints??