Microsoft Build 2023: Aye, I AI!

Top 5 announcements from Satya Nadella 

As always, watch the Keynote to get a taste of the full experience. Satya Nadella promised us 50 new announcements but only focused on his top five:

1. Bring Bing to ChatGPT - By grounding ChatGPT with a search engine, you are able to get results that are timely and more accurate.

2. Windows Copilot - ChatGPT on the desktop.

3. Plugins - The ability for third party vendors to extend ChatGPT by exposing their data through APIs.

4. Azure AI Studio - An end-to-end toolkit for building, training and deploying AI models including PromptFlow, a visual workflow tool.

5. Microsoft Fabric - The most ambitious and biggest announcement, Microsoft is                basically trying to create the Office Suite for Big Data - an integrated SaaS offering                that brings together all of your data in OneLake (a data lake using an open format),                then provides a consistent user interface for accessing that data regardless of your              role. I was able to use it in a lab in the afternoon and it definitely made it easier to                  bring together the disparate tools into a simple workflow.

And… the story of GPT

The other highlight for me was a fantastic session on the history of and the current state of GPT. Andrej Karpathy, AI Researcher & Founding Member, OpenAI, laid down a master class in 45 minutes.

Definitely check out the video for State of GPT, but here is the takeaway for you impatient types:

Do more things with Azure OpenAI

I am really excited to continue working with Azure OpenAI and using it to reduce busy work and mundane tasks. Or, to quote Microsoft CTO Kevin Scott:

When in Seattle, pound coffee. This latte came from the Grumpy Bean.

It says something about the current state of Microsoft when we don’t have a deep dive on new Windows features until the second hour of the second day keynote. Satya Nadella has really changed the focus of the company from a walled fortress to a…. well, open fortress. I mean, c’mon, this is Microsoft.

One of the most promising copilots has to be Microsoft 365 Copilot. According to research by Microsoft, nearly two days out of every five are spent in meetings or coordinating meetings. What if you had an AI assistant to help you manage that? Microsoft 365 Copilot combines ChatGPT and Microsoft Graph ( the API that allows you to access your entire Office suite as well as Sharepoint and Dynamics). Imagine being able to ask ChatGPT to “find and summarize my most important emails” after you come back from PTO and are facing 584 unread emails. Here’s the session.

Introducing Dev Home

To give developers a more integrated and simplified experience while they are developing, Dev Home has been introduced. For instance, developers can integrate their Github account with Windows Explorer and create a configuration for the development machine that can also be shared in Codespaces or dev box so they don’t have to worry about where to develop.

Tech people talking about tech

One of the nice things about being at a conference is the people you meet. Microsoft set up time to meet 1:1 with various executives to talk about targeted topics such as AI. I was lucky enough to get a chance to talk about the intersection of AI and video with Chris Pratley, Corporate Vice President at Microsoft. When we were talking about good use cases for Microsoft Copilot, he said that it was excellent at summarizing and gave several examples, such as using Microsoft 365 Copilot to summarize a Teams meeting that you joined late or had to leave early (or in my case, weren’t paying attention as it was going on). A really cool variation on that would be to use it to highlight points of interest in a longer presentation. For example, using Copilot to show you any part of the Microsoft Build Keynote where they talk about Microsoft Fabric.

Always go to a Mark Russinovich talk

Mark is the CTO of Azure and a legendary developer, so his talks are always packed with great tech. I attended Inside Azure Innovations featuring Mark Russinovich and I was not disappointed. In this talk, he covered:

• Hollow Core Fiber - Optic fiber cable with air in the middle that allows for higher bandwidth and lower latency
• Azure Linux - Yes. Microsoft has its own full Linux distribution. What!?
• Azure Container Storage - A pool of dedicated disks or SAN you can attach to your containers so there is no latency in allocating disk space for storage when you launch a container
• Dapr Workflow - Now you don’t have to create your own!
• Project Silica - Using glass as storage! It lasts longer than any hard drive or SSD. Just don’t drop it!

What’s Microsoft Build without a wrap party?

Finally, we ended the day with a party that included mini-golf, dueling pianos and a very large football table! But wait… there’s still Day #3!

The last day definitely had an after-party feel. It started late, ended too early, and, in the end, we were all forced out into the sun, blinking, stunned and unsure of where to go next.

But first! A latte and this beautiful view from inside the Seattle Convention Center:

Increase developer velocity with SQL

Some very cool things are coming out in SQL! You can watch the whole session, but for me, I was most excited about JSON becoming a first-class data type and the introduction of the Data API Builder. This tool automatically generates a REST API or a Graph API that accesses your data, automating the grunt work of creating an access layer for your data. The demo made it look easy (of course), so I’m looking forward to using it for real to see how it performs.

Enhance your solutions with new Azure AI products and features

This session gives a great rundown of all of the AI studio features and how to combine them to effectively create AI-enhanced experiences.

You definitely need to use the AI content safety feature, which can analyze both inputs and outputs of your AI to ensure that you are creating the right level of safety. It looks at the text for violence, self-harm, sexual or hate content. You can set the acceptable level - which of course would be different for an AI assisting in an online role-playing game for adults versus children.

The other great feature is PromptFlow: a visual designer that allows you to integrate plug-ins (APIs that can interact with Azure Open AI) and Azure Open AI.

Scott and Mark learn to code

If you want to learn a lot about how generative AI works while watching two programmers razz each other as they use GitHub Copilot to create a multiplayer version of Asteroids, then this is the session for you.

For example, they showed a great example of how temperature affects the output. Temperature is how “creative” the model is allowed to be. Low temperature tells the model to choose the next word with the highest probability and high temperature tells it to take more risks.

Here is an example of a prompt and response at a two, which is the highest (least deterministic) setting.

And here is the same prompt at zero temperature (most deterministic).

They go back and forth between using Github Copilot to generate code and then explain how it works, finally ending up with a working version of multiplayer asteroids.

and these great takeaways…

It was great to be there live! I met some interesting people and got to test drive Azure Open AI and Microsoft Fabric in a lab setting. Now, I just need to get Azure Open AI to summarize the sessions that I missed! Learn more about Kin + Carta’s Microsoft capabilities.