Yeah, that was built on the API for GPT-3.5, before even the chat API was introduced... I felt that given that OpenAI basically implemented the same thing inside their API/product (months later), there's no point in maintaining this anymore... But if you find this interesting, I could give it a go, and port it to the latest infrastructure.
You are ridiculously smart lol, you have images working before they do! I'm loving your clone, thank you very much. Would it be possible to add the ability to save the chat thread? Will it remember things I feed it if I were to use it to help write the story I've been working on in chat gpt? Also would it be possible to allow us to upload an image and have it give us the text prompts? Thanks again, IMPRESSIVE!
I know it's been 2 weeks since you left this comment, but we've been talking on the discord, so I didn't respond until now, but now been making sure I respond to all the comments. BTW, did you find out the "Clip Interrogator" (replicate.com/pharmapsychotic/clip-interrogator) and img2prompt (replicate.com/methexis-inc/img2prompt)? I shared them in my Midjourney videos, those tools allow you to give it an image, and it returns a prompt that allows you to generate a similar image.
Can you modify parameters such as topics for content that is considered illegal. Tried to use GPT to write content for a blog on cannabis but it wouldn’t allow it. When cloning is there a workaround to accept this topic?
The clone has no content it considers illegal. Just asked it to write a blog posts on the benefits of recreational cannabis use, and it wrote it easily.
I have buit several chatbot with different stacks such as vanilla, Next and Sveltekit, and I thnk I will go through this tutorial and dive in Nuxt fullstack applicaion because I am curious about Nuxt server route development, I think this will be a great tutorial 🤩😍😘😀
Great to hear! Let me know (either here on our Discord) how it goes, and what did you think! I'm also super curious about how you think it compares to the Sveltekit one.
Excellent video! could you do a follow-up using also embeddings so to finetune the answers to say a company internal target? I noticed that your prompt variable in the code could be populated with the 3 or 4 best probable embeddings coming from a vector store like Pinecone.
Yeah, this video is quite old in LLM terms, now there are so many other ways I could've built this. In this case, a vector store wouldn't help, since there isn't a "defined" knowledge base. But if I wanted it to answer data from a company data set or something like that, then I would definitely go with a vector store. I will make a video basically connecting Whisper/LLM/vector store to create a database of a full podcast.
Has anyone been able to deploy chatGPT clone to Netlify or Vercel using edge functions? That's where I stumble and I have yet to see anyone do it there and also get streaming to work.
I initially planned go deploy this to Netlify and Vercel, and couldn't get streaming to work. Something in the Vercel network blocks SSE. That's why I switched to Cloud Run, and took me a few minutes to get it to work, including streaming.
Ok I will definitely try! I did make a "vlog" out of my wifi lights video, not sure if that really worked out well, though. Would love to hear your opinion.