Of course, when we talk about the AI platform and the copilot stack, the next thing for us, which
is really exciting, is AI Studio. This is the full lifecycle toolchain for you to be able to build your
intelligent apps and your copilots, everything from being able to train your own models to be
able to then ground whether it’s OpenAI or any open source model with data that you bring,
built-in vector indexing in Azure Search, built-in support for RAG, or retrieval augmented
generation support, built-in support for prompt engineering with Prompt Flow and Orchestration,
and of course, built-in support for perhaps the most important feature, which is AI Safety.
One of the things that we have been hard at work is to build into the toolchain AI Safety. We’ve
been at work on AI Safety for the last five years. We have principles which we have translated
into a core set of processes that we implement across our engineering stack. And then, of course,
we have all of the compliance and oversight. But the real challenge is not just to have these
things outside the engineering process, but to build it into the everyday tool chain.
And that’s what we’re doing with AI Studio, and it starts with testing. There is the Responsible
AI dashboard that helps you during the testing phase to ensure that what you’re developing is
safe. We have grounding and, in fact, the Prompt Flow is perhaps one of the best features for you
to be able to ground your models. You have provenance, provenance for media, provenance
support for images and videos, and watermarking for your neural voice that’s going to be
available to all of you as you build your applications, and deployments in time.
That’s perhaps one of the most critical things, is we have taken all of the safety work we did, for
example, for the launch of Bing Chat, and really made it available as just a set of features for any
developer to use, right? You can take an OSS model and use the AI Safety service to really make
it, at the deployment time, safe. And of course, then you can even monitor the model for model
drift. And that way, then you can make sure that it’s not just a one time, but you’re continuously
looking to make sure that you have safe deployment.
We’re very, very excited about AI Studio helping every developer out here to be able to build AI
applications but build them with safety first. Let’s roll the video.
(Video segment.)
VOICEOVER: Introducing Azure AI Studio, a full lifecycle tool to build, customize, train,
evaluate and deploy the latest next generation models responsibly. With just a few clicks,
developers can ground AI models with their structured and unstructured data to quickly and
easily build customized, cutting edge conversational experiences for their customers.
Developers can take advantage of a new model catalog that works with the popular models
organizations use, including those from Azure OpenAI Service, Hugging Face and many other
open source models.
With Prompt Flow, developers can combine relevant data from your organization and create a
detailed prompts to get better results. Prompt Flow works with foundations, internally developed
for open source models, and uses popular open source tools, LangChain and Semantic Kernel.
And because the AI systems we build are designed to support our AI principles, with Azure AI