8/16/21 | By: Matthew Lallas & Stefano Paolina
Have you ever read an article or a book and not quite understood it? Have you ever had to read something and then summarize or give an analysis? Have you ever been in the midst of reading and part of your brain starts to predict what literary content is about to follow? GPT-3, a quickly growing AI application, is doing all of this right now. GPT-3 is a deep learning language processing model that allows AI to learn from a range of textual inputs and output a variety of results. These results can be anything from a summary of a Mark Twain book to an analysis of an MLK Jr. speech (transcript). GPT-3 can even take information - like surveys - and transform that into an easy-to-read and concise recap of the results.
OpenAI is a company quickly developing this type of AI and helping forward its use for hundreds of applications. For example, one company, Viable, uses it to better understand customer feedback. Open AI sources information from consumers in the form of surveys or comments feedback, and then the AI processes this information and outputs a concise summary of the information. This bypasses the need for companies to sift through hundreds and thousands of bits of information that will help them better their product or service.
Another implementation is by a subsidiary of OmnyLaunch, OmnyChat, which is using this technology to help the world become more verbally fluid and ease the struggles of learning a new language. OmnyChat offers a conversational environment that allows an individual to practice a new language. Imagine you want to learn Japanese. It is one thing to learn words, phrases, and sentences, but it is another thing to fluently communicate in that language; mastering colloquialism and authentic pronunciations - this is a whole other challenge. With the help of OmnyChat and AI, you can have a conversation in real-time in the language of your choosing. It would be the equivalent of having a conversation with a bi-lingual friend who is solely dedicated to helping you master the ins and outs of a language. Omnychat may even help progress your learning by exposing you to new ways of framing a sentence or a question. Ultimately, OmnyChat is bringing the language classroom conversations to your phone, computer, or internet device anywhere you are so your linguistic development never stops.
So how does this technology work? Without getting too deep in the weeds, GPT-3 used open A.I. technology to draw information from various pages throughout the vast expanse of the internet. Within 9 months of GPT-3's first commercial launch, more than 300 applications are currently using the software generating "an average of 4.5 billion words per day." https://openai.com/blog/gpt-3-apps/
GPT-3 is simple enough for anyone to use yet robust enough to handle industrial machine learning needs. A comprehensible parallel would be to compare GPT-3 to the text-suggest technology in your smartphone, but instead of predicting the next word, GPT-3 is predicting the next sentence, paragraph, page.
You can tell that nuance of literature are not entirely up to par, yet it is still able- in a matter of seconds - to give near-academic-grade quality content. The playground works by first giving the platform a prompt for example: "A Journalism editor was asked to edit and complete the unfinished essay below." The user can then control the length and the randomness ("temperature") of the response.
Although the technology is still in its infancy, there already lies a large amount of potential real-world use cases. Some of the most applicable use cases are teaching new languages, improving writing skills, as well as critical thinking. One of the more interesting features of GPT-3 is the ability to summarize a complicated piece of text and explaining it from someone else's perspective. One of the early demos involved having different characters summarize and explain the text, for example, if you input a textbook page on a problem faced in the Venture Capital space, GPT-3 could output a response explaining the text from Ben Horowitz's perspective.