May Habib (Writer): AI Primer on GPT-3

Blake: Writer is the AI writing assistant for Teams. One of the biggest developments of late in the AI writing world has been GPT-3. So what is GPT-3 and why are you excited about it?

May Habib: So, GPT- 3, specifically, is an AI model created to mimic human-generated language. The 3 is the model specifically built by OpenAI, who's commercialized it via an API. And to learn about human writing, GPT- 3, consumed 20 terabytes of text. To put that into context, that is all of Wikipedia, all of Google books, all of Common Crawl, which is everything published on the internet from 2016 to 2019. And very critically GPT-3 is a few shot learning, which in machine learning means, if you give it just a few examples of what you need, it can real really take that request and run with it. There are definitely a lot of... Well, maybe not a lot, but there are a number of open source GPT alternatives and more coming on the way. But really OpenAI's researchers, and what they've done on the commercialization piece is just the gold standard for GPT. Applications for GPT so far have been best for copywriting, product descriptions, Google ads, very specific formats. One thing that's not as well understood in all of the hype around GPT, some of it is hype, is just how many different types of writing there are. And copywriting specifically, ads and product copy, are really the things that current GPT apps are best at. So product descriptions for physical products, direct response style, Google ads, a lot of them for physical products. That's been where a lot of the revenue has been generated so far in the industry. For just about everything else, our tests have found that a human writer will have to edit the entire raw GPT output. Facts, but also just for plain sense. So it can sound very human, but not make sense in terms of how one idea flows into another. And right now the benefit of GPT is just really as a creativity booster, which can be great. You write a little, GPT writes a little, you revise what GPT wrote. You write a little bit more and by the time you are done, there may be nothing left of the GPT raw output, but you've written faster than you may have, if you are looking at a blank piece of paper. And that's really been what we've seen when we've put the various models to the test with real users and the real tasks they have at work. We are an AI writing assistant, and I am so excited about the way we are going to be integrating GPT technology in a writer. It's unique. It's never been done before. And we think we will finally enable people to use it across a wide range of their writing needs and styles at work. That's all I can say for now, but if folks are interested in joining us for the ride, we are hiring across every team and yeah, to build the leading AI writing assistant for professionals.