Why Microsoft’s GPT-3 license should be a win-win for all parties involved
Ever since its release, OpenAI’s GPT-3 series has persistently stayed at the centre of discussion among researchers, developers and entrepreneurs around the world. Regarded as one of the world’s most advanced machine learning-based text (and image) generators, it uses (in its full capacity), over 175 billion machine learning-based parameters to train its language representations – an increase of almost 11,500% compared to its predecessor (for reference, the entire English Wikipedia constitutes only 0.6% of GPT-3’s total training data.)
The sheer immensity of scale at which it functions, has made the GPT-3 one of the world’s costliest natural language processing algorithms, costing the firm a whopping US$12 million to research and train the final model. Additionally, the attached costs also involve (1) tens of thousands of dollars in monthly cloud computing or server and electricity costs for running the model; (2) possibly more than a million dollars in yearly retraining costs due to model decay; and (3) additional costs of customer support, marketing, IT, security, legal and other requirements of running a product. This could be in the tens of thousands of dollars based on the number and size of customers OpenAI acquires.
Financing the GPT-3
In the era of commercial AI, research labs like OpenAI usually require deep pockets from wealthy tech firms in order to finance their research over the long term. It was in 2019 that OpenAI made the transition from a non-profit organisation to a for-profit company in order to cover its costs in their long quest for the development of artificial general intelligence (AGI). It was also then that Microsoft came in with a US$1 billion investment into the firm.
Given its steep training costs and invaluable potential, OpenAI faces a great challenge in even breaking even on the costs of developing, training and running its huge GPT-3 neural network, let alone turning it into a revenue-generating business. It should thus come as no surprise that it has decided to commercialise its landmark natural language processor with a carefully crafted slabbed-payment scheme along the lines of a SaaS (Software-as-a-Service) model, instead of making it open-source. This has been done through its tie-up with Microsoft, who in September 2020 declared that it would be exclusively licensing the GPT-3 platform as a part of its ongoing partnership with OpenAI.
Given the cost structure for the GPT-3 and its pricing plan, it is expected that OpenAI will require several dozens of customers opting for its Build Tier plan (US$400/month) just to break even on running costs. Additionally, it will require much more to cover the expenses involved the development, training and (eventual) retraining of the model.
The issue in this, however, is the fact that there is no precedent to this. GPT-3 is the first-of-its-kind natural language processor capable of both zero- and one-shot learning (in image processing, for example, instead of assuming image generation as a classification problem that requires massive training data, GPT-3 turns it into a difference-evaluation problem, supplying images with similarity rating in proximity to the image being trained). Hence, finding the appropriate use case for businesses becomes a challenge.
Historically, many (possibly useful) products have died swift deaths because they could not find the right product-market fit that allowed them to acquire new customers in a cost-effective way. So far, however, OpenAI claims to have received several thousands of applications to run its GPT-3 model (on a free beta version) for businesses to find an appropriate application for the software model. But what will be crucial to OpenAI in making the business model successful over the medium run, is their licensing relationship with Microsoft.
The Microsoft Factor
“Microsoft is teaming up with OpenAI to exclusively license GPT-3, allowing us to leverage its technical innovations to develop and deliver advanced AI solutions for our customers, as well as create new solutions that harness the amazing power of advanced natural language generation.”
–Kevin Scott, Chief Technology Officer, Microsoft.
The point to note here is Microsoft’s customer base. Over a million companies globally use the MS Office suite and the digital assistant Cortana. Additionally, Bing is the world’s second-most popular search engine. These are technologies which make prodigious use of natural language processing technology – and the adoption of the advanced GPT-3 engine could be exactly what both parties need, moving ahead.
For OpenAI, the hundreds of millions of users will provide key insights and experience that will allow GPT-3 to grow and adapt. For Microsoft, they will be the exclusive owner of one of the world’s most advanced AI technologies. Seems like a win-win, if there ever was one.