Microsoft is having a hell of a week. On Monday, the company announced a $US7.5 ($10) billion deal to acquire Zenimax and all of its big video game properties. In one move, the Xbox platform’s future suddenly looked a lot brighter. Today, Microsoft announced another deal that could also have a huge long-term impact in the coming years: It’s acquired an exclusive licence for OpenAI’s GPT-3 language model.
OpenAI has been making headlines for years — primarily due to the fact that AI-phobe Elon Musk was an early investor. But more recently, OpenAI’s autoregressive language models have been able to claim attention on their own, and other big-time investors have come forward. Specifically, Microsoft pumped a billion dollars into the secretive company last year. This morning, that relationship got even cozier with Microsoft’s announcement that it is exclusively licensing the GPT-3 language model for an undisclosed sum.
We live in an era when lofty claims of AI capability are often hyperbolic to the point of dishonesty. GPT-3 is not the world’s first computer brain. But it is a pretty convincing language model that outperforms anything we’ve seen when it comes to mimicking human-like speech and taking straightforward instruction in plain English. Microsoft sees the power here, and it’s boxing out the competition. (By competition, we mean Amazon.)
Here's a sentence describing what Google's home page should look and here's GPT-3 generating the code for it nearly perfectly. pic.twitter.com/m49hoKiEpR
— Sharif Shameem (@sharifshameem) July 15, 2020
Kevin Scott, Microsoft’s executive vice president and chief technology officer, wrote in a statement that he’s excited to leverage GPT-3’s “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.”
Scott was vague about what “solutions” the company has in mind but emphasised the value that GPT-3 will have for Microsoft, only saying that it will use the model’s capabilities “in our own products, services, and experiences to benefit our customers” and that OpenAI’s work offers benefits in “human creativity and ingenuity in areas like writing and composition, describing and summarizing large blocks of long-form data (including code), converting natural language to another language.”
The announcement isn’t totally clear what the licence covers. We asked Microsoft for more details and a spokesperson told us that “an exclusive licence gives Microsoft unique access to the code behind the GPT-3 model, which includes a set of technical advancements, so that they can rapidly and tightly integrate capabilities directly into Microsoft’s products and services, thus delivering new and powerful AI-powered solutions to their customers.”
Scott wrote that developers will continue to have access to OpenAI’s closed API, which will run exclusively on Microsoft’s Azure cloud service. He specifically cited “researchers, entrepreneurs, hobbyists, businesses” as examples of groups that will still be able to use the model. In a statement, OpenAI also made it clear that “the deal has no impact on continued access to the GPT-3 model through OpenAI’s API.” We’ve reached out to Microsoft for further details on what areas of exclusivity it’s laying claim to but we did not receive an immediate response.
As for use cases, we can probably expect Microsoft to focus on boring enterprise solutions that won’t ever be relevant to the average consumer. But GPT-3 could be used for content generation, automated customer service, chatbots, simple website design, translation services, voice assistant optimisation, conducting job interviews, and yes, it could even write articles like the one you’re currently reading.
As Scott wrote in his announcement, “we haven’t even imagined” most of GPT-3’s capabilities. That unknown has caused a lot of handwringing about the potential for abuse of the system as well as inherent biases that can likely be found in all machine learning products.
When OpenAI first showed off GPT-2, its previous model in the series, it openly expressed worries that it was too dangerous to release in full to the public. For perspective on how far the model has come, just know that GPT-2 had 1.5 billion parameters while GPT-3 is working with 175 billion parameters. The latest model is quite the leap forward and theoretically that much more of a risk for those worried that it could be weaponised into a disinformation machine the likes of which we’ve never seen.
GPT-3 still has a long way to go before it could reliably communicate with and deceive humans about its synthetic origins. But anyone who’s taken a look at message boards for the increasingly-influential Qanon conspiracy theory knows that online disinformation doesn’t need to contain clarity or coherence to have a massive impact.