Research group OpenAI announced today that it will sell access to its cutting-edge language generation tools to clients with the organization’s first commercial product.
The move comes a week after the group unveiled GPT-3, its most sophisticated text-generating machine learning system yet. GPT-3 was trained on a dataset nearly 100 times the size of GPT-2, its previous groundbreaking effort from 2019.
While the full code for GPT-2 was initially withheld out of fear that it would be put to nefarious use, the nonprofit-for-profit hybrid group eventually made the complete version free through open source in fall of last year. OpenAI seems to be taking a different tack with GPT-3, however, allowing API access to invitation-only clients with pricing undisclosed.
So far, clients include Reddit, Quizlet, mental health platform Koko and a company called Replika, which develops “AI companions.” Those companies are using it for applications such as search functions that can better understand context and meaning, sophisticated chatbots and content moderation.
OpenAI claimed in a blog post that the commercial release will better control how its technology is being used as well as generating revenue to fund its research efforts. Because of its size and the computing power necessary, GPT-3 alone cost $12 million to train.
“The field’s pace of progress means that there are frequently surprising new applications of AI, both positive and negative,” a spokesperson wrote. “We will terminate API access for obviously harmful use cases, such as harassment, spam, radicalization or astroturfing.”
Originally founded by Elon Musk in 2015 with the mission to bring the benefits of AI to the broader good of humanity, OpenAI caused some controversy last year when it announced it was moving from a nonprofit model to a hybrid for-profit last year in order to attract investors. It announced soon after that Microsoft would be investing $1 billion in its research.
With its first commercial product, OpenAI seems to have weighed the benefits of continuing its goal to make AI more accessible against the expenses needed to do so.
“The API has pushed us to sharpen our focus on general-purpose AI technology—advancing the technology, making it usable, and considering its impacts in the real world,” the spokesperson wrote. “We hope that the API will greatly lower the barrier to producing beneficial AI-powered products, resulting in tools and services that are hard to imagine today.”
OpenAI’s GPT-2 helped spur a new research boom in natural language processing technology but until now, its commercial applications seemed to be limited by the unpredictability of the output. Even GPT-3 continues to suffer from some of these problems, a disclaimer section in the 73-page paper OpenAI recently put out on the system described a tendency toward racist and sexist bias in some instances.
Correction: An earlier version of this article listed Wikimedia Foundation as an API client. It is not at this time.