One of the biggest trends in machine learning right now is text generation. The AI system absorbs and learns billions of words scraped from the Internet and generates text in response to various prompts. It sounds simple, but these machines can do a variety of things, from creating fiction to writing bad code to allowing you to talk to historical figures.
The most well-known AI text generator is OpenAI’s GPT-3, which the company recently announced, and it currently has “tens of thousands” of developers in use in more than 300 different apps and produces 4.5 billion words per day. that’s swallow Robot horse. While this could be any milestone that OpenAI can celebrate, it is also a useful indicator of the growth of the scale, impact and commercial potential of AI text generation.
OpenAI started out as a non-profit organization, but over the years it has been trying to monetize the use of GPT-3 as its first marketable product. The company has an exclusive agreement with Microsoft and gives this tech giant unique access to the program’s base code, but any company can apply for access to GPT-3’s generic API and build services on top of it.
OpenAI is very interested in advertising, so now hundreds of companies are doing exactly that. A startup named Viable uses GPT-3 to analyze customer feedback and identify “subjects such as surveys, help desk tickets, live chat logs, reviews, sentiments, and sentiments.” Fable Studio is using this program to create conversations for VR experiences. Algolia is using it to improve their web search products and sell them to other customers.
All of this is good news for OpenAI (and Microsoft, whose Azure cloud computing platform supports OpenAI’s technology), but not everyone in the startup space is enthusiastic. Many analysts have pointed out that it is stupid to build a company based on technology that they don’t actually own. Creating a startup using GPT-3 is incredibly simple, but it’s incredibly simple for its competitors as well. There are ways to differentiate GPT startups through branding and UI, but none of them can gain by using technology as much as OpenAI.
Another concern about the rise of text generation systems is related to print quality issues. Like many algorithms, text generators can absorb and amplify harmful biases. They are also often surprisingly dumb. In testing of a medical chatbot built using GPT-3, the model responded to a “suicide” patient who prompted suicide. These problems cannot be overcome, but it is worth reporting in a world where algorithms are already imposing false arrests, unfair school grades, and biased medical bills.
But, as OpenAI’s latest milestone shows, GPT-3 just keeps talking and needs to be prepared for a world full of robot-generated conversations.