What Is Retrieval-Augmented Generation and How to Make AI Work for You, with Guil Hernandez

🎓 View our courses: https://scrimba.com/links/all-courses 🎙 About the episode Meet Guil Hernandez 🇺🇸! He is a developer and educator with over 15 years of experience in tech. He's also a Scrimba teacher who is a part of the team bringing you the AI Engineer Path, and in this episode, he's helping us understand retrieval-augmented generation. In the previous episode, Tom Chant helped us understand the world of AI models. Today, Guil will further teach us how these models work under the hood. AI models don't understand the world like we do. When we interact with them, they turn our inputs into mathematical representations known as embeddings. By creating our own embeddings, we can teach AI to do what we want it to. Today, we're getting an introduction about making a model aware of your own data source so that that data can be considered for the AI output. For example, using the techniques you'll learn from Gil in this episode, you could connect a model to your customer support conversations so that the model knows what is necessary to answer unique questions about your (or your client's) business. This is the third episode of our series on AI engineering, introducing Scrimba's AI Engineer Path. This path is your gateway to unlocking the full potential of AI for your projects. ⏰ Timestamps Guil focuses on RAG and embeddings (01:42) RAG makes a foundation model aware of your data (03:14) Spotify has been using RAG since 2014 (05:56) How embedding works: embedding model + vector database + generative model (09:00) You're enhancing content retrieved from a database with a generative model (10:26) A foundation model can't just understand text (10:34) What's a vector database? (12:35) Can we make an AI chatbot for the Scrimba podcast? (15:05) You can chunk the files directly at OpenAI now! (16:49) OpenAI's Assistants API (17:33) AI is evolving quickly (19:07) Assistants API does RAG (19:55) What is fine-tuning? (20:39) Differences between RAG and fine-tuning (