Conversational Technologies
7 min read

How Businesses Can Leverage GPT-3 and The OpenAI API

The Hyro Team The Latest Hyro News & Updates
How Businesses Can Leverage GPT-3 and The OpenAI API

‍When AI researcher and game designer Jason Rohrer created a chatbot using OpenAI’s text-generating language model GPT-3 for fun, he likely didn’t anticipate the endeavor ending in an outright ban from using the company’s technology.

 

Rohrer named his chatbot creation “Samantha” and programmed her to be friendly and curious. He allowed others to customize his creation, cryptically dubbed Project December, to build their own customized chatbots, with one man notably creating a close proxy of his deceased fiancée. 

 

Eventually, OpenAI learned about the project and gave Rohrer an ultimatum: modify it to prevent misuse and implement an automated monitoring tool, or shut it down entirely. Rohrer refused, and, soon after, he was informed by OpenAI that he was no longer allowed to use GPT-3.

 

While the above scenario presents an interesting ethical dilemma, it’s evident that GPT-3 is a uniquely powerful program that can be applied in a number of different settings. Let’s explore the basics of GPT-3, the OpenAI API, and how the next generation of machine-driven language models can impact businesses of all kinds.

A Brief Overview of GPT-3 

Generative Pre-trained Transformer (GPT-3) is a machine learning-driven language model developed by the OpenAI artificial intelligence lab. Put simply, GPT-3 generates human-like text using pre-trained algorithms.

 

To understand what this means here’s a quick test: How would you complete the following sentence? “I wanted to make French toast, so I went to the fridge and got some ______”.

 

Chances are, you said “eggs”, “butter”, or “milk”, as, given your extensive knowledge of delicious breakfast foods, those are the most logical items needed in this scenario. But, while it may be simple for a human being to fill in the blank in a coherent way, this task is not as straightforward for most machines — until now.

 

A program like GPT-3 is essentially a statistical model that uses neural networks and machine learning to calculate the conditional probability of words, or how likely a word is to appear in a sentence given the other words that precede it. So, in the example above, “eggs”, “milk”, or “butter” are statistically more likely to follow than say, “rutabaga” or “battery acid”.

 

GPT-3 isn’t the only language model out there, but what sets it apart from others is its sheer size. It’s been fed a colossal amount of information from all around the public web including Common Crawl (a repository of web crawl data filtered for quality), the entirety of Wikipedia, and several other coding and math databases, as well as other datasets from the OpenAI lab. As a result, it’s capable of generating various types of text, including scripts, poems, guitar tabs, code, and even essays.

What is the OpenAI API?

The commercial GPT-3 based product is called OpenAI API. At present, there are over 300 applications, tens of thousands of developers, and more than 4.5 billion words per day generated with the API. Essentially, users can input any text prompt (like a sentence or paragraph), and the model will generate a response in fluent, natural language.

 

The Case for an API Over Open Source

 

When it comes to GPT-3, some people have quipped that OpenAI should consider rebranding to ClosedAI. While OpenAI did share their algorithms with the public for previous iterations like GPT-2, they’ve decided to take a different approach and keep the algorithm under wraps for the time being.

 

But why? The answers are manifold.

 

One reason is financial in nature. In 2019, Microsoft announced they were investing $1 billion into OpenAI as part of a multi-year partnership. This move granted Microsoft exclusive source code access to GPT-3, allowing them to bypass the API that every other organization or individual must use. Many people view this decision as OpenAI moving away from their original values of “democratizing AI”, something that OpenAI vehemently denies.

 

OpenAI argues that putting the source code behind a paywall allows them to monetize its research, as, without deliberate and carefully managed experimentation and testing, GPT-3 cannot advance in the direction they want it to. They also argue that the code is simply too complex for most people to run and the API provides a much more approachable route to engage with the program.

 

The OpenAI team argues that another reason to limit access to GPT-3 is to protect the general public from bad actors who could easily co-opt the AI for more nefarious purposes. It’s disturbingly easy to see how a tool that has been trained on writing by humans could be easily swayed to spread political misinformation or hate speech online, or even to impersonate real people without their consent.

The OpenAI API in Business

The Impact of Natural Language Generation

 

One specific subset of AI that is set to make an impact in the business world is the field of Natural Language Generation (NLG), a domain responsible for converting structured data into meaningful phrases in the form of natural language.

 

Today, we generate a colossal amount of data and the quantity continues to grow every year. In fact, between 2018 and 2025, the size of real-time data in the global datasphere is predicted to expand tenfold. When it comes to packaging and displaying this data, there are few mediums more powerful than language. Sure, a dashboard can communicate data with attractive visuals, but a few sentences can tell a story and bring that data to life, which is precisely why natural language conversational AI is so highly sought after in businesses today. All companies need to generate reports but doing so can be tedious and time-consuming. If time-intensive reports and data analysis could be taken care of by AI solution, employees would be freed up to focus on more creative or fulfilling tasks.

 

GPT-3 has been called “the next generation of NLG” due to its ability to understand data, extract meaning, and identify relationships between data points that can then be communicated in plain English. Let’s take a closer look at some of the other ways GPT-3 based apps are being used to shape the world.

Parsing Unstructured Data

 

Unstructured data is precisely what it sounds like — data that isn’t stored in a structured database format. Some examples of unstructured data include geo-spatial data, audio, photos, text files, and more. Other types of data might fall somewhere in between. For example, email can be considered semi-structured because it includes structured data like the date, time, and sender and recipient names, while the body of the email doesn’t follow a specific format.

 

So, why is parsing unstructured data so important? According to one study from CIO.com, up to 90% of the data we generate is unstructured, and this type of data is growing by a staggering 55-65% every year. That’s a vast amount of data that isn’t particularly useful until it can be structured, to allow businesses to collate, analyze, and gain insights from it.

 

GPT-3 can be applied to parse unstructured data and allow users to create tables from long-form text. Essentially, the user feeds in a text prompt, specifies the desired structure for the data and provides some examples. In turn, the system will return a table of results with the same structure every time.

 

Open AI’s parser provides an example of how this works. The image below shows a sample prompt that consists of text detailing the types of fruit found on the fictional planet Goocrux.

(Source)

 

Given this prompt, here is what the sample response looks like:

Improving Written English

 

As a language model, GPT-3 is committed to continually improving its command of the English language. However, this isn’t a conscious endeavor.

 

Much like the way we don’t consciously learn a language as a child, neither does GPT-3. If someone were to ask you for tips on how to learn your native language, you’d probably struggle to come up with anything useful and practical — you were simply fed the language until you picked it up. GPT-3 works in much the same way.

 

GPT-3 is now so competent in its mastery of the English language that it can help individuals improve their English and produce high-quality materials. For example, the Grammar Correction app on OpenAI can correct errors and suggest alternative versions that are more aligned with Standard English. There’s also an app that can interpret complex text and produce a more simplified, accessible version, as well as an app that can summarize long pieces of text into more concise passages (a TL;DR generator, if you will).

OpenAI API’s Strengths and Weaknesses for Businesses

While many AI systems are designed with a single-use in mind, the OpenAI API operates on a general “text in, text out” basis, rendering it an all-purpose API. However, that’s not to say that the API can’t be used to create tools for specific use cases. Developers can use the API to create apps for customer service, chatbots, and productivity, as well as tools for content creation, document searching, and more, with many providing great utility for businesses. For example, there’s an app that can detect sentiment in Tweets which is useful for companies looking to gain insight into the public perception of their brand, as well as an app that can turn a simple product description into functional promotional ad copy.

 

Where the API fails businesses in its utility is in the area of fresh and relevant information. Although GPT-3 has been trained with a gargantuan amount of data, it’s not always great at assigning weight to the most up-to-date and relevant information. This creates a situation where a tool built on the API might be able to name every plant native to Bali, but struggle to name a newly elected official if that event occurred recently.

 

Businesses are forever developing and evolving. If nothing else, the pandemic has demonstrated just how quickly things can change and how rapidly new information becomes stale. That’s why for enterprises that rely on scaling with real-time data, dedicated conversational AI systems are more effective at providing their audiences with engaging, timely, and helpful conversational experiences.

 

All companies have a distinct culture, brand language, processes, and data. In order for an AI solution to produce meaningful results, it must be trained using business-specific parameters. GPT-3 began as an all-purpose, “anything goes” AI super-experiment — it wasn’t explicitly designed to allow companies to supercharge their data. As a result, companies can get much more value by instead opting for conversational AI solutions that can handle industry and company-specific knowledge in real-time.

About the author
The Hyro Team The Latest Hyro News & Updates

Updates from Hyro's Marketing team.