OpenAI has released two major tools for developers to make use of GPT-3 and DALL-E. GPT-3 is the radical new text generation tool, which generates large or small amount of texts from simple prompts. It can also classify text into categories GPT-3’s text-generation system forms the core of OpenAI’s new chatbot, ChatGPT. DALL-E is an image generation tool, which creates images from text prompts. Together, these two tools provide today’s state-of-the-art in AI-based content generation.
So how are developers making use of these new features? Today we are looking at several ways that these tools have been put to use.
Basic Content Generation
The core of GPT-3 is generating content from prompts. Whether for making blog posts, writing summaries, or finding bullet points in text, many tools have popped up which essentially just feed queries to GPT-3 and give the results back to users. Many of these make use of templates which help users write better prompts for GPT-3. For instance, the content creation tool Jasper has a number of templates, including a blog post template. Using the template will ask you for the topic, the tone, the audience, and other important details, and then use those to construct a query to GPT-3. This sort of “simple wrapper” around the GPT-3 engine dominates the current crop of GPT-3 apps.
Several companies have made use of GPT-3s ability to summarize data in unique ways. Detangle.ai, while not officially released yet, aims to summarize legal documents in a human readable form. Eightify will summarize long YouTube videos and deliver relevant points. OpenAI themselves are working on tools to summarize entire books for you.
Just-in-Time Game Content Creation
One of the keys to working with AI well is knowing its limitations, and when those limitations don’t matter as much. While it might be problematic to use GPT-3 to write an instruction manual for air conditioner maintenance, there are other areas where slight inconsistencies actually won’t matter at all, such as game content creation. Fable Studio is working to make the chat feature of GPT-3 to help power their online characters, allowing users of the game to interactively speak with the characters in a much more direct and natural way.
One game actually tries to make an entire adventure game based on GPT-3 prompts, called AI Dungeon. This game was an interesting proof of concept, but it also shows some limitations on how far you can take this concept. I can see in the future having games where many NPCs (non-player characters) have backstories that are generated by GPT-3, and perhaps images generated by DALL-E, allowing for new stories every time you play the game. One might even have a GPT-3-generated world, but it will require the game makers to spend some more time taming GPT-3 with non-AI code to make it work well.
Of course, DALL-E itself provides images based on prompts, and there are plenty of tools that wrap this feature similar to the way that the GPT-3 content creation does. However, there are also some design-based tools that are more creative in their use of OpenAI’s tools.
Jordan Singer developed a tools which takes a description of a website and generates a mockup of it in the website design tool Figma. Danny Richman developed a program which, given a prompt, will generate a color palette which matches your description.
AI has the ability to make customer service representatives much more efficient. ActiveChat.ai and Sapling will auto-generate subject-specific content and inquiry responses for customer service representatives. GPT-3 provides automatic content creation for customer service representatives, who then only have to verify the data before sending it back to customers.
On the other side of the ball, ChatDesk will summarize the contents of customer service interactions so that managers have better visibility into the general tenor of customer service inquiries at a glance.
Another interesting use case for GPT-3 is as a suggestion engine. If you give GPT-3 sufficient prompting and constraints, it will provide suggestions for just about anything. To start wiith, PlaylistAI is a tool which generates a playlist for you based on a description. You could describe a movie scene, a mood, an attitude, and this tool will generate a playlist to match. Friday is a shopping assistant. Just type what you are interested in and see what pops up! For movie and book recommendations, check out Serendipity.
As you can see, developers around the world are coming up with interesting uses of OpenAI’s tools. Which of these will take off and which ones will be duds? Only time will tell.