John Seabrook’s “The Next Word” (https://bit.ly/2xrKh7j) is the best thing I’ve read this week. Its an article on predictive text and it tries to find if an A.I writer that is trained on the archives of the magazine, can learn to write for The New Yorker. But what makes this article special is that at the end of each section, readers can see what the New Yorker trained A.I. predicted would come next. In collaboration with OpenAI, an artificial-intelligence company, The New Yorker was able to use their newly released A.I writer, called GPT-2 to test its capabilities with predictive text.
GPT-2 is just one of OpenAI’s numerous A.I projects, They even have a gaming A.I that beat the world’s best human team at Dota 2, a multiplayer online strategy game. They also developed software for a robotic hand that can teach itself to manipulate objects of different shapes and sizes without any human programming. In this article, at the end of each section, there is a small piece of text that was predicted to come after the previous paragraph. GPT-2 is able to predict text in a very human-like manner and it’s eerily similar to John Seabrook’s style.
Towards the end of the article, there is a section that allows us the choice where we think the author’s writing ends and the AI’s begins and it’s pretty hard to choose the right answer as they are both indistinguisable. This article is quite long but goes into depth about the development steps and intricacies of GPT-1’s A.I architecture and I would definitely recommend this to anyone who is remotely interested in technology.
Steve's BLOG. 