I'm your new Subscriber ❤. Can you make a long video that how we can develop those AI application from scratch to end and, also hosting... 🎉 Thanks you🎉
yeah the fundamental feature of sitegpt is actually to just be able to chat with any website, that's what I was trying to rebuild. They go further with it because they also allow you to import the chatbot into your own site so customer can chat with it, I didn't build that part, I just wanted to show how you can build a quick prototype with the core functionality
Bro ,I have a question : If we want to create a robust app like this we should know how gpt and AI work ?,what's the API keys, how does the internet work to deploy the web site,etc.... but if we take the code from gpt I think it's not a good idea especially for beginners ? Can you suggest me your opinion for this because the majority they feel overwhelmed
When you are just starting up, use chatgpt more like a tutor. So you don't straight up copy/paste code from it, you just use it to guide you step by step and explain each step to you so you can do it yourself. Imagine literally having a senior software engineer next to you who can guide you, he won't tell you everything to write, but he will point you to the right solution. So to use it efficiently, give it your code and ask it to optimize it, ask it to explain how it would build feature X or Y, etc. That's how you will learn fast
Asalam alaykum, brother. I recently put together an autogen framework where the different LLMs take a response and then they work on the code, but instead of restating the whole code block, they say this "insert bad code with errors" needs to be replaced with this "insert new code" and it dynamically creates an outside python script that changes the specific code that was quoted (line number col etc.) in the python file that the LLMs are working on. So it is really cheap because it saves on tokens (if they have a low context window). That way, it is extremely easy, fast, and cheap to fix small bugs with small models, without letting the models make something up (since they're a small model) and change the overall code in ways that don't make sense, inshaAllah. After the syntax errors are fixed, all the bugs in the PROBLEMS section that can't be fixed by simply typing new code (library issues or import issues) will be passed to the next bigger model to have that model answer how those errors should be addressed. After that, it is passed to the next bigger model, which will reccommend specific next steps. After that, the largest model (claude 3.5 inshaAllah) will evaluate all their work. And i forgot to say in the beginning that the user inputs the idea for the code they want and the claude model will output the full thing and then pass it to the smallest model for quick fixes and so on. If you want the code, i can send it inshaAllah. Btw, i wanted to say remember Surah Kahf 23-24 inshaAllah Also, inshaAllah, remember Abu Darda' reported Allah's Apostle (ﷺ) as saying: If anyone learns by heart the first ten verses of the Surah al-Kahf, he will be protected from the Dajjal. (Sahih Muslim 809a)
The goal was to show how you can quickly build the core functionality of many AI Saas in a few hours, I originally did it for the students of my bootcamp. Ofc it's not complete (it's missing Stripe integration and several other things), these are meant to be MVPs
Yep check the new video I released, I am gonna do more detailed breakdowns in my next videos🤝 Also you can access the source code for all 7 apps on my X account, its in the pinned tweet
How many AI wrapper SaaS apps have been made with this same stack? I've seen 4 different yt dev channels explain the same thing, and these people rarely seem like they know what they're talking about. Making it a point to say "I use useState and useEffect." just makes you sound like those are the only hooks you know how/when to use. (And you are probably using useEffect wrong if that's the case) Based on what I've heard in interviews, these types of AI wrappers are prob gonna be completely worthless when newer gen models with better multimodality & cheaper inference costs come out later in the year or early in the next. Maybe taking software dev seriously would be better use of time instead of doing this Marc Lou roleplay.
Brother the stack I am talking about is to build personal projects, not production ready SaaS apps. This is the stack I used to build 7 AI projects in 7 weeks. I documented the whole process on my X account with links to the apps live, the source code, and everything. You can find it there. About "useState and useEffect", yeah for these projects, they cover the majority of what I need from React. Once someone understands how to use them, they can build most stuff they want. After that it's more about optimization, build custom hooks using them, etc. AI wrappers is a different story. It's not as simple. ChatGPT has integrated a chat with pdf feature a while ago, and yet these "Chat with PDF" wrapper apps still make a bunch of money even today (check pdf.ai - that guy makes something like 60k/mo, it's a solo dev). But many will get killed I agree. I think the only once who will survive are those that have good distribution and user engagement will stay. Distribution leverage is key here. The guy from pdf.ai is doing well cause he was one of the first in the market, has good domain, + hired influencers for tiktok marketing (so distribution).
WaAlaykum Assalam brother! How would you talk to it then when you are chatting with AI? To my knowledge there is nothing wrong, there are even references to companions using "you" on objects, like when Omar (RA) talked to the black stone saying its just a stone.
@@thecodebendermaster That is a very good point. I typically just make sure to refer to it in the third person by the name of the bot. The reason i avoid saying you is because from the robot's "perspective" it would be internalizing that as "he is referring to me" since they can technically "talk" or put out text. If there is a machine that can put out text like humans language, that's fine because it is all just a big mathematical translation of words into vectors and we know that. But if we also attach a kind of perspective to the robot that is able to put out language, it is close to imitating the creation of Allah. People like to say they have consciousness, and that's only because the robot can be trained to say certain things that sound human and an object saying "i" is definitely something that is unique to humans. You brought up a good point with Omar raddiAllahu anhu tho
This is good i feel like in this year onwards having knowledge on how AI works, And how to use it is going to be CRUCIAL for a developers arsenal, especially to get a major head start against people that are not using it. I also feel like i learn way more and especially more efficiently when learning coding. Very nice video
alternatively you could also get it for free if you use an open source mode like Llama, several sites have free credits they give you like the NVIDIA Nim API
HI sir, I'm shubhankit from India, I am so eager to learn from you but I don't have 500 dollars I have only 50 dollars to pay you. Watching you since a year and collecting money to get into your course. Please sir is it be possible to get into your course somehow????
Hey man! Send me an email to zaurbek.stark@lastcodebender.com with your resume and explain why you want to join the course. Depending on that, we will try to figure out something, I offer scholarships to a few students. Also fyi I am out this month so I will be slow to answer, I will be more responsive in July
It's the internal AI API, I am not using any LLM locally here. With nextjs you can specify API routes that your frontend will call, those routes will be like middlewares between your app and OpenAI. The way I use it is: my frontend -> calls internal API -> this adds to the prompt + passes the OpenAI API key -> we use it to make a call to OpenAI Hope this helps bro, lmk if I can clarify something!
I am waiting for a video card order to come in so I can use an uncensored AI LLM locally for security research (chatgpt drives me crazy concerning hacking related stuff, a lot of which is censored, so you have to "trick" it all the time to get proper results).
