Microsoft recently launched a ground-breaking feature on its Azure OpenAI service. It now allows enterprises to feed their proprietary data directly into GPT-4 or ChatGPT.
Known as “Azure OpenAI on your data”, the new functionality takes away the need for enterprises to fine-tune and train their own generative AI models. It’s currently available as a public preview through the Azure OpenAI Service.
Andy Beatman, the senior product marketing manager for Azure, called the new feature a “highly requested customer capability”.
He further emphasized its potential in harvesting valuable customer insights, monetizing access to data, and gaining deep industry and competitor insights.
A Brief Overview of the Feature: How Does It Work?
Azure OpenAI service now supports connecting to several sources, including Azure Cognitive Search, Azure Blob storage container, and local files. With its new capability, Azure OpenAI Service can retrieve and connect the necessary data from any source.
When a user runs a query on Azure, Microsoft’s cloud service identifies the internal corporate data necessary to complete it. The data is then retrieved and combined with the original query to create a new query.
Microsoft Azure OpenAI passes on the new query to the AI model of choice. The AI model runs the query and delivers the result, which is then sent back to the user.
Microsoft explained that together with Azure Cognitive Search, Azure OpenAI on your data determines what data it needs to retrieve based on not only the user’s input but also the available conversation history.
Microsoft’s Focus on Proprietary Data
Microsoft has already invested over $10 billion into OpenAI. The tech giant is now integrating the startup’s AI tools and models into products and services across its vast portfolio at a rapid pace.
Microsoft’s focus on exploring ways to craft tailored AI models capable of generating more specific results is pretty evident. These models aren’t restricted by their training – they are customized for specific applications and organizations.
Vendors have started to get on board with it in recent months as generative AI innovation picked up pace.
The approach isn’t exactly new – it has been a topic of discussion for years. Later last year, Nvidia rolled out NeMo. A framework built within its larger AI platform – it enables organizations to enhance their LLMs using proprietary data.
During the lead-up to Nvidia’s GTC 2023 show in March, the company’s vice president of enterprise computing Manuvir Das told journalists that many enterprise companies are “interested in creating models for their own purposes with their own data”.
The GPU giant collaborated with ServiceNow a couple of months after the event. The team-up allowed customers using Nvidia AI tools and ServiceNow’s cloud platform to train their own AI models.
Caveats of The New Functionality
Microsoft Azure OpenAI for your data comprises several shortcomings. Each model response has a max token limit of 1,500, which includes the user’s query, system messages, internal prompts, retrieved search documents, and the response.
Hence, users are required to break down long questions into several small ones.
The functionality might potentially give rise to a major security concern by leaking corporate secrets into the public domain by training its AI models.
Not long ago, more than 100,000 compromised accounts were found for sale on the dark web. Since ChatGPT stores the records for every conversation by default, this might also leave sensitive proprietary information vulnerable.
Still, Microsoft Azure OpenAI for your data appears to be quite promising. It remains to be seen whether the new functionality will find wide adoption among large enterprises despite its flaws and potential risks.