Microsoft has a hell of a week. On Monday, the company announced a $ 7.5 billion deal to acquire Zenimax and all of its major video games. With one move, the future of the Xbox platform suddenly looks much brighter. Microsoft today announced another deal that could also have a huge long-term impact in the coming years: It has acquired an exclusive OpenAI license GPT-3 language model.
OpenAI has been making titles for years – mostly due to the fact that AI-fob Elon Musk was an early investor. But lately, OpenAI autoregression language models have managed to demand attention themselves, and other large investors emerged. In particular, Microsoft is pumping billion dollars in secret company last year. This morning, this connection became even more cozy with that of Microsoft message that it exclusively licenses the GPT-3 language model for an undisclosed amount.
We live in an age when high claims of AI ability are often hyperbolic to the point of dishonesty. GPT-3 is no the world’s first computer brain. But this is a quite convincing a language model that surpasses anything we’ve seen when it comes to imitating human-like speech and making direct instructions in plain English. Microsoft sees the power here, and this is boxing for the race. (By competition we mean Amazon.)
Kevin Scott, Microsoft’sxecutive vice president and craise ttechnology officer, writes in declaration that it is excited to use GPT-3’s “technical innovations” to develop and deliver advanced artificial intelligence solutions to our customers, as well as to create new solutions that harness the incredible power of advanced natural language generation. “
Scott was unclear about the “solutions” the company had in mind, but stressed the value that GPT-3 would have for Microsoft, saying only that it would use the model’s capabilities “in our own products, services and experience to benefit our customers “And that OpenAI’s work offers advantages in” human creativity and ingenuity in areas such as writing and composition, describing and summarizing large blocks of data in long forms (including code), converting natural language to another language. “
The message is not entirely clear what the license covers. We asked Microsoft for more details and a spokesman told us that “the exclusive license gives Microsoft unique access to the code behind the GPT-3, which includes a set of technical enhancements so they can quickly and tightly integrate capabilities directly into Microsoft products and services. , thus providing its customers with new and powerful solutions powered by artificial intelligence. “
Scott writes that developers will continue to have access to OpenAI closed API,, which will run exclusively on the Microsoft Azure cloud service. He specifically cites “researchers, entrepreneurs, amateurs, businesses” as examples of groups that will still be able to use the model. IN statement, OpenAI also indicated that “the transaction has no impact on continuous access to the GPT-3 through the OpenAI API.” We contacted Microsoft for further details on the areas of exclusivity it claims, but did not receive an immediate response.
In terms of usage, we can probably expect Microsoft to focus on boring enterprise solutions that will never matter to the average consumer. But GPT-3 can be used for content generation, automated customer service, chat bots, simplified website design, translation services, voice assistant optimization, job interviews and yes, you can you even write articles like the one you are currently reading.
As Scott wrote in his message, “we didn’t even imagine” most of the capabilities of GPT-3. This unknown has caused many handshakes about the potential for abuse of the system, as well as the inherent biases that are likely to to be found in all machine learning products.
When OpenAI first showed GPT-2, its previous model in the series, it openly expressed concern that it was too dangerous to be released entirely to the public. For the perspective of how far the model has come, just know that the GPT-2 had 1.5 billion parameters while GPT-3 works with 175 billion parameters. The latter model is quite a leap forward and theoretically a much greater risk for those who are worried about it could be armed in a disinformation machine like we’ve never seen before.
GPT-3 is still there long way to go before he can reliably communicate and mislead people about his synthetic origins. But anyone who has looked at message boards for the increasingly influential Qanon conspiracy theory knows that online disinformation does not have to contain clarity or coherence in order to have a huge impact.