Event banner
Event details
Hello everyone, and thank you to the Microsoft team for this informative session. I'm John Pearce, Founder at Ainnovétion, where we're actively using Azure AI Studio to develop various models for our AI-powered healthcare solution. As we look to incorporate a generative AI component into our software for 2024, we are considering Azure OpenAI for this purpose. Our specific need is for a model that can automatically pull and incorporate new data points for each individual patient to provide accurate and personalized responses. Could you shed light on whether Azure OpenAI models support this kind of dynamic data ingestion and updating? Additionally, how does Azure OpenAI compare with other platforms in terms of integration, scalability, and security, particularly in a healthcare application context? We're eager to understand the best path forward for our project and greatly appreciate your insights on these critical aspects.
- CourtneyBrewer
Microsoft
You might be interested to know – if you haven’t found them already - Azure AI Studio includes a gallery of industry-specific prompt samples helps facilitate the development of domain-specific copilots. Whether developing enterprise chat solutions, enhancing customer interactions with multimodal experiences, or delving into speech analytics.- I appreciate your introduction to the prompt samples in Azure AI Studio, I was not aware of this. I'm particularly intrigued by the recent announcement of Phi-2. Its performance, achieving state-of-the-art results among base language models with less than 13 billion parameters, is impressive. Phi-2's ability to match or outperform models up to 25 times larger on complex benchmarks, is particularly noteworthy. Given its compact size and outstanding reasoning and language understanding capabilities, I'm considering Phi-2 as a foundation for our initial AI projects in healthcare. Are there any particular considerations or advice you'd recommend regarding its use in this context? I understand that Phi-2 is designed as an ideal platform for research and development, including safety improvements and fine-tuning for various tasks. I'm eager to hear about any applications or use cases of Phi-2, particularly those that might align with our AI healthcare solutions. Insights into how Phi-2 is being utilized in different scenarios, considering its tailored data curation and scaling techniques, would be extremely valuable as we explore its suitability for our projects.
- SharmichockYou can finetune phi-2 on your own dataset to align the model to the new domain for best results.
- deeikeleIf you’re looking to use Azure OpenAI service to personalize the patient’s response, this is typically well served by a pattern we call “Retrieval augmented generation” (RAG). With this you pass along additional information about the patient as input to the large language model. In Azure AI studio, take a look at playground for prototyping a chat based application, or prompt flow to iterate and evaluate your RAG flow.
