SQLBits 2024
Navigating the Skies of Open AI: Perfect Your Grounding for a Smooth Landing
Explore the evolution of knowledge consumption with ChatGPT in this insightful session. We'll dive into the importance of "grounding" in Large Language Model (LLM) applications and introduce you to the emerging world of vector databases. While delving into OpenAI's technology, participants will gain an understanding of vector embeddings, scalability challenges, and efficient data retrieval methods. Designed for data scientists, engineers, and decision-makers, this Level 300 session offers a balanced overview for both technical and non-technical attendees.
The advent of ChatGPT has revolutionised the way we consume knowledge across the internet. This session will explore how OpenAI's technology can similarly unlock your internal company knowledge. To achieve this, we delve into a technique called "grounding," which is as crucial to the success of Large Language Model (LLM)-powered applications as executing a perfect landing is for a pilot.
As SQL and database experts, you're probably already well-versed in traditional databases. LLM-powered applications introduce a new type of database—vector databases—that you'll need to understand to stay ahead of the curve.
What will we cover?
• Review the impact of ChatGPT and how to harness OpenAI's technology for internal knowledge access.
• Gain insights into the Retrieval Augmented Generation (RAG) pattern and how it contrasts with techniques like fine-tuning.
• Learn the fundamentals of vector embeddings and their role in accessing unstructured data.
• Explore emerging Vector Database solutions, with a focus on Microsoft Azure Cognitive Search Vector DB.
• Delve into scalability considerations, exploring the limitations of vector embeddings and how vector databases are addressing these challenges.
• Acquire efficient grounding techniques, including data chunking and prompt engineering, to enhance data retrieval and response quality.
• A brief look at alternatives to vector databases, such as using Graph DBs or hybrid search solutions.
• Consider additional factors such as cost minimisation, adapting to new LLM model versions, and the trade-offs between smaller LLM models and GPT-4.
• A short demo featuring Azure OpenAI and Azure AI Search for creating a RAG based application.
Pre-requirements
This Level 300 session is tailored for data scientists, engineers, and decision-makers navigating the evolving landscape of generative AI. No prior knowledge is required, the session will begin with high-level introductions before delving into more detailed concepts, aiming to engage both technical and less technical audience members.
As SQL and database experts, you're probably already well-versed in traditional databases. LLM-powered applications introduce a new type of database—vector databases—that you'll need to understand to stay ahead of the curve.
What will we cover?
• Review the impact of ChatGPT and how to harness OpenAI's technology for internal knowledge access.
• Gain insights into the Retrieval Augmented Generation (RAG) pattern and how it contrasts with techniques like fine-tuning.
• Learn the fundamentals of vector embeddings and their role in accessing unstructured data.
• Explore emerging Vector Database solutions, with a focus on Microsoft Azure Cognitive Search Vector DB.
• Delve into scalability considerations, exploring the limitations of vector embeddings and how vector databases are addressing these challenges.
• Acquire efficient grounding techniques, including data chunking and prompt engineering, to enhance data retrieval and response quality.
• A brief look at alternatives to vector databases, such as using Graph DBs or hybrid search solutions.
• Consider additional factors such as cost minimisation, adapting to new LLM model versions, and the trade-offs between smaller LLM models and GPT-4.
• A short demo featuring Azure OpenAI and Azure AI Search for creating a RAG based application.
Pre-requirements
This Level 300 session is tailored for data scientists, engineers, and decision-makers navigating the evolving landscape of generative AI. No prior knowledge is required, the session will begin with high-level introductions before delving into more detailed concepts, aiming to engage both technical and less technical audience members.