This article is part of our series that explores the business of artificial intelligence
Developers will soon be able to use GPT-3, OpenAI’s flagship language model, through the new Azure OpenAI Service, announced at the Microsoft Ignite conference this week.
The public release of GPT-3 comes after one year of limited trials through OpenAI’s own API service and a few specialized integrations with Microsoft’s software development products.
With the release of Azure OpenAI Service, Microsoft will make the power of large language models available to a wide range of organizations and industries. It will be an opportunity for the software giant to strengthen its hold on the new business applications taking shape around advances in natural language processing and generation.
The new service can also have important implications for OpenAI, which is becoming increasingly dependent and entrenched in the business goals of Microsoft.
Moving GPT-3 to production level
“We are just in the beginning stages of figuring out what the power and potential of GPT-3 is, which is what makes it so interesting,” said Eric Boyd, Microsoft corporate vice president for Azure AI. “Now we are taking what OpenAI has released and making it available with all the enterprise promises that businesses need to move into production [emphasis mine].”
Since the release of GPT-3, many developers have used it to create interesting and sometimes unexpected applications. From email composition to text summary, website design, and even software code generation, the large language model has shown promise to be a platform for innovation and new products.
But there’s a big difference between proving that a deep learning model works and creating a working business model on top of it. Companies that want to build products with technologies like GPT-3 must overcome various technical, legal, ethical, and financial hurdles.
These are the challenges that Microsoft is addressing with the rollout of Azure OpenAI Services.
Very large neural networks, like the one powering GPT-3, have very complicated computational requirements, making it difficult to run them cost-effectively. Without an optimized and scalable infrastructure, the costs of running GPT-3 would be very high, and those costs would be projected onto the price of products offered to customers. The rise in price would in turn reduce the size of the prodcut’s total addressable market. Since the beginning of its partnership with OpenAI, Microsoft has engaged in a series of projects to create computational hardware that are tailored to the needs of deep learning models such as GPT-3. This allows Microsoft to offer GPT-3 at an affordable price to businesses of different sizes.
Microsoft is also taking care of other technical problems such as security, access management, private networking, and data handling protections. This will further expand the market by reducing the technical burden of product teams, especially for small startups that don’t have the in-house talent to address these challenges.
According to Azure’s website, the difference between OpenAI Service and the OpenAI API is that “OpenAI Service brings together OpenAI API and Azure enterprise-level security, compliance, and regional availability.”
Microsoft will also monitor for fair use, bias, and other ethical considerations that have arisen around large language models such as GPT-3.
Integration with Microsoft products
GPT-3 is an enabling technology. Alone, it does not provide much value beside showing how far advances in deep learning have come. To get real value out of it, you must either use it to build a new product, create a market for developers to build products on top of it, or integrate it into existing products and services.
When launching a new product, you must find its target market and establish efficient channels for customer acquisition. Whether you’ll be attracting customers through ads, conferences, direct sales, or free product trials, if the average cost of acquiring a customer is higher than the value generated by that customer, then you don’t have a working business plan unless you can live on investor money until you reach profitability.
You also need to have “moats,” or protection mechanisms that ensure customers stay on your product. If another developer can create a copy of your product and outmaneuver you with a bigger marketing/sales budget or more competitive prices, then your product won’t have a promising outlook. This is why launching a successful standalone GPT-3 product is very difficult: You’re either going to fail because you can’t turn in profits or you’ll be eliminated by wealthy competitors who can outspend you and even run their product at a loss until they dominate the market.
Interestingly, while GPT-3 is often touted as a technology that democratizes language models, its real beneficiaries are the incumbents who already have a working product and a sizeable market.
This is where Microsoft is using its strengths to its advantage. Microsoft can use its exclusive license to OpenAI’s technologies to launch new GPT-3-powered products at low costs. But its real strength is in the two other areas: the developer market and product integration paths.
Microsoft owns a vast array of software development tools as well as GitHub, the largest online source-code management platform. Through these channels, Microsoft can immediately offer its GPT-3-powered development tools to millions of developers without spending much on finding marketing channels.
Through GitHub, Microsoft released CoPilot, a specialized version of GPT-3 that can generate source code snippets. It also released another GPT-3-powered deep learning model that generates queries for Power Apps, a low-code development platform for business applications. Microsoft also owns Visual Studio Code, one of the most popular integrated development environments (IDE), which supports deep learning–powered source code generators. GPT-3 has already made its mark in the software development community, and according to GitHub, for some languages, 30 percent of newly written code is being suggested by Copilot.
Equally important is Microsoft’s reach in non-tech sectors. More than a billion people use Word, Excel, Outlook, Teams, and other productivity, communication, and business tools developed by Microsoft. And Microsoft is already exploring how it can provide GPT-3-powered features to those applications. Integration provides a much shorter path for Microsoft to create value and differentiation through GPT-3.
Microsoft’s market reach and its direct access to GPT-3 also puts it in a unique position to gather new data and find new ways to improve its products, hardware, and AI models to squeeze more profit out of OpenAI’s technology and solidify its spot as a leader in AI innovation.
OpenAI began as a non-profit scientific AI laboratory seeking a safe path to artificial general intelligence. It later transitioned to a for-profit organization seeking funding for its research. Its relationship with Microsoft began in 2019 with a billion-dollar investment that had some strings attached.
Since then, the ties between OpenAI and Microsoft have grown stronger. Today, OpenAI’s technology has become an integral part of Microsoft’s AI strategy. On the other hand, Microsoft has become virtually the only channel through which OpenAI can monetize its technology and fund its research (I wouldn’t be surprised if OpenAI API gradually phases out).
How will this affect OpenAI? Already, the company is being criticized for no longer releasing the source code and parameters of its deep learning models and protecting them like commercial companies protect their intellectual property. Will the partnership with Microsoft develop into an acquisition (as is Microsoft’s habit)? Will it steer OpenAI’s research away from AGI and toward products that have short-term business value? Time will tell, but the stories of other AI and robotics research labs show that maintaining the balance between business goals and scientific endeavors is very difficult.