I built a ChatGPT @lex_fridman plugin that allows to chat with the collective wisdom contained across 499 of Lex's podcasts and videos... Here's how I built it 🧵

#ChatGPT #ArtificialIntelligence #nlproc #OPStack

Started by:
- downloading Lex MP3s via @YouTube API
- transcribing audio to text with @OpenAI's large Whisper model
- (optional) hosted the new "lex-transcripts" dataset on @huggingface datasets!

‼️ The code for the next few steps can be found here:
github.com/pinecone-io/ex…

Next, I used @LangChainAI to process the podcasts and break the text down into chunks of ~400 tokens

To make these podcast chunks searchable, I embedded them using @OpenAI's text-embedding-ada-002 model and stored them in the @pinecone vector DB

To integrate this with ChatGPT we need an API that ChatGPT can call. @OpenAI's pre-built "template" plugin was the perfect starting point. I forked it and made a few changes to create the ask-lex-plugin 🛠️
github.com/jamescalam/ask…

Next, I hosted the repo via @digitalocean — now everything was ready for ChatGPT...

Over on ChatGPT, I created the plugin via "Develop your own plugin". I installed it, and just like magic ✨ it was ready to use and giving incredibly detailed answers

If you want to develop your own, I created an in-depth video on how this was built here:

I can also share access to the plugin with a small number of people that have access to ChatGPT plugins — DM me if interested