What you missed at Microsoft Ignite 2023

Microsoft announced that Fabric, its SaaS offering for Big Data, is now generally available. The platform provides a unified interface for all users, improves Power BI performance through Direct Lake, simplifies data lake setup through One Lake, and allows you to access files on AWS as if they were on Azure - among many other features. If you haven’t looked at it yet, you should.

Microsoft is all in on Copilots. Copilot is the AI chat orchestration framework that simplifies Generative AI deployment by integrating hooks for pre and post-request filtering, allowing you to set prompts and check for incorrect or objectionable responses. Some of the announcements were rebranding (Bing Chat is now Copilot) and brand simplification, while other announcements were significant, such as:

• Copilot Studio, a low-code tool for creating your own Copilot
• Azure OpenAI is adopting the OpenAI plugin standard that will allow you to extend the Copilots by grounding them in your own data via APIs
• Copilot for Power BI and Fabric

A live coding exercise with Github Copilot provided a lot of fun and useful advice. Scott Hanselmann and Mark Downie did a fantastic job of explaining exactly what GitHub Copilot can do, how it can help you, and why you should always check its work! They really drove home the fact that it is a non-deterministic prediction machine that needs context to be accurate and a human to guide it and validate its results.

Another great day at Microsoft Ignite - here are my top takeaways from Day 2.

Microsoft has combined its rapidly changing and ever-expanding Generative AI offerings with the rest of its AI services into one cohesive workspace. Azure AI Studio is an all-in-one AI platform for building, evaluating and deploying generative AI solutions and custom copilots. Now in public preview, it allows you to deploy different versions of your model (Microsoft, Meta, Open AI, open source providers or your own custom model), test them against standard datasets or upload your own, then compare the results.

Copilot for Azure - in public preview - uses large language models (LLMs) to understand and interact with Azure resources and services. It can access data from Azure Monitor, Resource Graph, Application Insights and more and can be used to execute tasks (“find all resources without a ‘Cost Center’ tag) or to create reusable templates and scripts (“Create a script that finds all resources without a ‘Cost Center’ tag).

More about Microsoft Fabric… one of the great features of Fabric is that it doesn’t require you to move all of your data to a central location. However, it does benefit from having all of your data nearby. With mirroring, you get all of the deliciousness of lower latency with none of the messiness of having to set up a data pipeline. Once you enable mirroring for Snowflake, Azure Cosmos DB, or Azure SQL DB, it will create corresponding parquet files in One Lake and keep them in sync.

Microsoft definitely saved the best for last: All of the above plus a double shot of Mark Russinovich (Azure CTO).

When Brendan Burns (co-inventor of Kubernetes) and Mark Russinovich (Azure CTO) team up to promote something, it is going to be impressive. The Radius platform hopes to solve one of the problems in DevOps where everyone has to know everything. How many skilled developers are also excellent network engineers and infrastructure engineers? How many operators are great coders? Some, but not all.

Radius was created to provide an abstraction layer via templating and recipes that will allow greater reusability. It is designed to be cross-platform and can currently deploy to Azure, AWS, and your local box. Watch the Radius session to see it in context.

The more I see the Azure AI studio, the more I like it. This session showed how to use Azure AI Studio and prompt flow to create and run complex AI interactions, and then monitor them for groundedness, relevance, and coherence over time. I was excited to see the code-first view of prompt flow and the prompt flow CLI interface in action. Get started with prompt flow here. Even more interesting was seeing a live demo where Gen AI was used to fix problems with Gen AI outputs.

While it has been pretty cool to work with all of the text generation and summarization that ChatGPT provides, I was even more excited to see the new model, GPT-4 Turbo with Vision (GPT-4V), that utilizes both natural language processing (NLP) and visual understanding to analyze images and provide text-based answers to related questions. This could allow you to easily add visual search to your website. Some examples:

• Have the user take a picture of the part they are looking for. GPT-4V can find it and answer questions about what it sees
• Have GPT-4V monitor a video feed for anomalies
• Use GPT-4V to convert and summarize pages of handwritten documents into text

All this and more! Check out the session for more information and some great demos.

For my booster shot of Russinovich, I sat in on the final keynote of the day where Mark goes through all of the amazing hardware and software that is used to run Azure. From innovations in cooling that will help reduce the carbon footprint of data centers to software proxies that allow global scheduling of AI workloads, he covered it all.

Even if you are not an infrastructure nerd, I encourage you to watch it so you can gain an understanding of and an appreciation for Azure and what an amazing engineering feat it is. And, of course, there is the obligatory game of Pong played on 11.9 terabytes of memory and 1700+ processors. You are looking at CPU activity being spun up and down to create a Pong emulator.

I expected to get a lot of AI in every session, but I hadn’t expected it to be so in-depth, practical, and applicable to my job. Thanks, Microsoft Ignite! I’ll see you next year!