What is OpenAI GPT-3?


The age of commercial AI, available as a cloud API is here. One can call an AI API, like any other Cloud service, and build products around it.

One of the biggest players in this space is OpenAI. Founders are already starting & building profitable companies, built around products enabled by OpenAI APIs.

The possibilities of building new products or integrating OpenAI into existing products are limitless. And this goes beyond for-profit products.

This guide aims to help Founders, Developers, Product Managers understand what it takes to build products with OpenAI GPT-3 API

What is OpenAI GPT-3

In layman’s terms, GPT-3 is an AI program that generates, classifies, or summarizes human-like text when prompted with a few examples. GPT in GPT-3 stands for Generative Pre-trained Transformer and 3 means it’s a third model in the n series of GPT AI models.

It’s created by OpenAI, an AI company that started as a non-profit founded by Elon Musk of Tesla, SpaceX, Sam Altman from YC, and others.

What makes OpenAI GPT-3 special?

GPT-3 is a few-shot learner model. Being a few-shot learner model means given only a few examples, GPT-3 will respond to text prompts with text completion or classification very close to the expected result from a Human. For example,

OpenAI GPT-3 Playground Example

In this example, GPT-3 was prompted with just the first three inputs. Based on such a limited amount of example inputs, it understood that it’s being asked to generate

  1. A list, consisting of
  2. SuperHero Characters
  3. Only from the Marvel Universe

This makes it very useful to use in the real world without requiring a lot of training data. OpenAI has made it easier by hosting the models themselves and exposing the prompt and the output part through a simple API.

The Opportunity for Application Developers and Startups

This has enabled a big opportunity in the democratization of the ability to create AI-enabled end-user products. Till a couple of years back, using ML/AI in your product meant training gigantic models with millions and currently billions of parameters, which frankly is not possible for 99% of the people and companies. You could build and train small models but they were not of much use in an end-user product.

With the availability of GPT-3 as an API for example in the text generation and classification field, a single developer can build apps without having the expertise or the computing resources to train expensive AI models. This opens up building AI-enabled apps for a huge range of creators.