My Highlights from AI related Announcements at Microsoft //Build 2024

While most of you have now seen the announcements about the general availability of GPT-4 in Azure OpenAI, here is the list of my highlights from yesterday’s AI announcements that you may have missed:

1. New Models Added to the Phi-3 Family on Microsoft Azure

Microsoft announced the addition of new models to the Phi-3 family, now available on Microsoft Azure. The Phi-3-small and Phi-3-medium models, tailored for generative AI applications requiring strong reasoning, limited compute, and low latency, are now accessible to developers. Additionally, the previously available Phi-3-mini and Phi-3-medium models can now be utilized through Azure AI’s models as a service offering, enabling quick and easy integration. Learn more.

2. Azure AI Search Enhancements

Azure AI Search now supports AI Vision multimodal and AI Studio embedding models. These updates include integrated vectorization for multimodal search capabilities, handling both text and images during indexing and querying. The service also supports new embedding models from Azure AI Studio and Azure OpenAI, simplifying the integration process for vector or hybrid search and enriching retrieval augmented generation (RAG) applications. These updates are part of REST API 2024-05-01 Preview version. Details here.

3. Cost-Effective RAG at Scale with Azure AI Search

Several improvements to Azure AI Search were announced, significantly increasing storage capacity and vector index size at no additional cost. These enhancements position Azure AI Search as one of the most cost-effective solutions on the market. The service now supports vector search quantization and narrow numeric types for vectors, with options to reduce vector field storage utilization. Explore further.

4. Asynchronous Content Filters in Azure OpenAI Service

Microsoft introduced Asynchronous Filter in the Azure OpenAI Service. This feature allows for faster streaming experiences by running content filters asynchronously, meaning content is returned immediately without buffering. While this improves latency, it comes with a trade-off against real-time vetting of smaller sections of output, potentially allowing some harmful content to be displayed before being flagged. More information.

5. Azure API Center: Comprehensive API Inventory and Governance Solution

Azure API Center is now generally available, providing a centralized API inventory for seamless discovery, reuse, and governance of APIs. The service supports a complete and current catalog of APIs across various types and stages, enabling API best practices at scale and fostering API reusability. Features include built-in metadata, integration with CI/CD pipelines, and enhanced search and filter capabilities within the Azure portal. Check it out.

Conclusion

Build 2024 showcased Microsoft’s ongoing commitment to innovation, offering powerful new tools and enhancements across AI, cloud computing, and API management. These advancements highlight how Microsoft is shaping the future of technology, making it more accessible and efficient for developers and businesses worldwide.

About the Author

Aaron (Ari) Bornstein is an AI researcher with a passion for history, engaging with new technologies and computational medicine. As Applied Machine Learning and Data Science lead for Microsoft Healthcare in Israel, he works on developing disruptive technologies for the Healthcare industry using advanced NLP algorithms. Previously he worked in positions in which collaborated with the Machine Learning Community and Israeli Start Up Ecosystem. In 2017, Ari founded Olim In Tech, a volunteer led community geared towards empowering & connecting Olim who work or want to be working in hi-tech.