Build your own ChatGPT with GPT-4 and Vercel AI in 10 minutes! Stream responses and custom prompts!
Вставка
- Опубліковано 25 лип 2024
- Build a ChatGPT clone in mere minutes with these powerful new libraries. Streaming responses and custom prompts!
📰 Newsletter 📰
Sign up for my newsletter, "A Bit of SaaS," to keep you informed, inspired, and engaged with the latest trends, tips, and resources in the ever-evolving SaaS industry.
ethanmick.com/newsletter/
📜 Code & Docs 📜
SOURCE CODE
• ethanmick.com/build-your-own-...
Docs:
• sdk.vercel.ai/docs
• platform.openai.com/account/a...
👨🏼💻 About Me 👨🏼💻
I am a principal software engineer and architect who loves building cool cloud software.
• Website: ethanmick.com/
• GitHub: github.com/ethanmick
• Twitter: / ethan_mick
• LinkedIn: / ethan-mick
• Agency: ethanmick.dev/
🔖 Chapters 🔖
00:00 - Intro
00:29 - API Route
04:37 - Client Side
09:46 - Custom Prompts
13:25 - Outro - Наука та технологія
Best video on this topic ❤❤❤
Can you make a video of Chat with pdf using vercel AI sdk and Opensource models
@ethan_mick Awesome video! Do you have a video or advice how to hook this simple app into proprietory data sets and repsond to answers based on information in the data sets?
Great video! I'm so glad I found your channel. Do you have a style guide you use? I see you use eslint but do you also use a formatter? Thanks :)
Yep, I use Prettier (prettier.io/) for all my formatting. I mostly use the defaults with some small tweaks: No semicolons, Spaces, and 2 space indents.
Awesome stuff, man. Exactly what I needed to know!
Thanks!
Awesome video, thanks for sharing. Any insight as to how well this works in production when deploying on Vercel? I implemented something similar for a pet project in Node but I'm wondering if I could have just gone with NextJS for my backend too. Specifically, what if the stream takes a long time (say 45 seconds) to finish, does that cause issues? Edit: to be more specific, I ask the question because of Vercel's serverless/stateless nature.
Specifically the `edge` runtime makes this work really well on Vercel.
vercel.com/docs/concepts/limits/overview
The edge runtime let's you stream for... forever, actually:
> Edge Functions provide an alternative that do not have a maximum execution timeout and that also allow you to continue streaming responses.
So when you setup your app to run on the edge network it'll stream the entire response.
You're an amazing communicator and teacher. You should consider creating a Discord server asap!
Thanks! It's on the to-do list, just don't want it to consume too much energy.
LFG. We ready for that AI colab yet? I have some great plans haha. I’ll email you today. Awesome work lately man!!
I have some sweet content planned for playing with the Llama 2 and open source models. These models let you tweak them behind the scenes without needing to use OpenAI, it's awesome
hey, great tutorial, kindly, i have one question, if i want GPT to respond based on the different json files i have; like 2 different json files, how can i work that out with this tutorial?
You can do it one of two ways:
1. Put it in the prompt. Say, "If the JSON contains XYZ, then do this. Otherwise, do that."
2. Make two System prompts and check which JSON is which, and use the appropriate prompt. If you know the exact structure of your JSON, this way will give you better results.
@@ethan_mick Thank you for your response, i will try that out,
Hlo bro i am facing a problem on vercel. In last month i deploy my project on vercel without make git repo then now i want to download my project source code. Please tell how i download project source code
Hey, you could give this a try: github.com/vercel/vercel/discussions/5783
(No affiliation, just found it with some searching).
how do we force it to only answer questions in the scope given? and also not respond with the generic As an AI ......
Welcome to the world of Prompt Engineering!!
Basically, you can change the System Role to give your AI a whole new persona. Tell it to only answer as a persona, and to never say they are an AI. For example, I've turned GPT-4 into a medieval farmer for a game and if I try to jailbreak they call me crazy and end the conversation. It's awesome.
@@ethan_mick all of this is kind of new to me, do you have any tutorials on a process like that?
Thanks in Advance
@@404Thoughts I'll add a Prompt Engineering video to my to-do list, but I don't have one yet. Stay tuned!
I always say everywhere, don't use openai api, don't help training their systems. Corpos which using bad human rights situation in poorest countries to collect biometrics for fake cryptomoney is obviously evil corpo.
Do you have an alternative you suggest?
ua-cam.com/video/ax_R4yz1WwM/v-deo.html