At the same time, despite their obvious strengths, these models suffer from certain limitations, which can make it difficult to integrate them directly into our custom business workflows or applications. In particular they are confined to the data they were originally trained, so they would not know about specialized domains or our own internal data, and they are prone to hallucination (“making stuff up”).

However, it's possible to extend their reasoning to any custom, business- or domain-specific data set, opening avenues for more personalized, insightful and reliable integrations.
This session focuses on GPT models from OpenAI and explores the essential architectural components within Azure that support such integrations, outlining how to use these models in our own products and services.