Top 10 use cases of GPT-3 that are already out there Unsplash
- GPT-3 was developed by Elon Musk-owned OpenAI and is currently the largest artificial intelligence language model.
- It goes beyond just autocomplete to do things, bring historical figures back to life, create fanfiction all on its own, or even file a complaint against your landlord using legal language.
- OpenAI allowed certain developers access to the model’s private beta and here 10 mind-blowing use cases that they’ve already come with.
GPT-3 may be the largest artificial intelligence language model ever but it can do a lot more than just autocomplete your phrases.
The team at OpenAI began drip-feeding the software to select developers who requested access to a private beta. And, they’ve been busy coming up with some really interesting use cases for GPT-3.
For now, OpenAI wants third parties to explore what the GPT-3 is capable of. In the long run, the company has plans to turn the tool into a
commercial product that businesses can subscribe to and access via the cloud.
Here are 10 things that GPT-3 can do that go way beyond just autocomplete:
1. A question-based search engine
@paraschopra/Twitter
It’s like Google Search except in Q&A format. You ask the engine a question that it points you in the right direction to find your answer, like a Wikipedia page.
2. A chatbot that lets you talk to historical figures
Wikimedia
GPT-3 has a whole host of textbooks, biographies and autobiographies at its disposal. This means that, with a little modification, GPT-3 can be configured to a particular personality.
So you can ask Elon Musk about rockets, learn writing from Shakespeare or get
lessons on philosophy from Aristotle.
3. Creating conversations between people who have never met
Pxfuel
Fanfiction can be at your fingertips with GPT-3. Just as with historical figures, GPT-3 can create conversations between people who have lived across different eras, places and may never even have crossed paths.
One user created a meeting between Alan Turing and Claude Shannon. Another added what that same conversation could turn into if the imaginary characters from JK Rowling’s Harry Potter — including Harry himself, Hagrid and Hermoine — suddenly showed up.
4. Turn normal words into something fancy
Unsplash
GPT-3 can take normal text and give it a different style. For instance, you can ask the model to change a complaint against your landlord into ‘legal language’. You can do the same for emails, turning the tone from informal to formal, or even write up essays in the desired format — persuasive, descriptive or narrative depending on your homework assignment.
5. It can make its own music
Unsplash
Other music may not be so readily available, but guitar tabs are posted on the internet as ASCII text. That means GPT-3’s got some of the greatest hits stored away in its database. It has a few chords to start off with, the program can write its own music. The verdict is still pending on whether or not it ‘sounds good’.
6. Write creative fiction
Wikimedia
7. A doctor with endless patience
Unsplash
GPT-3 can be used to answer medical queries as well. Not only will the model offer a diagnosis of the issue, but it will explain the underlying issues along with the possible types of treatment.
8. A new era of text-based games
Screenshot of AI dungeon
Text-based games were one of the first to appear in the digital world back in the 1960s. They may not be the most popular in today’s day and age but among the top titles today, AI Dungeon, is powered by GPT-3.
9. Generate code based on text descriptions
@sharifshameem/Twitter
The GPT-3 can spew out code if you can tell it what you want. For instance, ask for a div — section of code — that contains three buttons, each with a random colour and you’ll get the relevant programming language.
10. GPT-3 can autocomplete images too
OpenAI
Just as its predecessor, GPT-2, the same underlying architecture of GPT-3 can be configured for pixels instead of words.
This means that just as it can autocomplete sentences, it can also autocomplete images if it has enough data.