A Challenger Approaches! Andrej Karpathy’s recent GPT Tokenizer video came with a fun request, create a companion guide/blog post of the already made video automatically.
In this video, we go over my approach to this, and how I used LLMs like GPT-4-Turbo, Audio Models like Whisper-1, and Embedding Models from OpenAI with vector databases to fully automate this process.
Github Page for Code & MD File: github.com/ALucek/companion-g...
@AndrejKarpathy Tokenizer Video: • Let's build the GPT To...
Andrej’s Tweet: / 1760740503614836917
Semantic Chunking: python.langchain.com/docs/mod...
Chapters:
00:00 - Intro & Context
01:46 - My Solution!
03:48 - Process Overview
04:43 - Downloading & Chunking Audio
07:23 - Transcribing with Whisper-1 & Post Processing
09:57 - Semantic Chunking Transcript
12:08 - LLM Prompting & Setup
16:04 - Initial LLM Output Overview
16:45 - Embedding Transcript for Similarity Search
17:46 - Searching for & Inserting Links + Pictures
20:12 - Main Script Overview
21:15 - Cost, Time, & Token Consumption
21:49 - Revisiting the Markdown Document
22:54 - Limitations & Drawbacks of My Approach
26:00 - Outro
Негізгі бет Ғылым және технология Turn Videos Into Blog Posts With AI! - GPT-4, Whisper-1, and Embedding Model Approach
Пікірлер: 10