GPT-3: Your New Lifelong Companion

I've been in the OpenAI API beta for a few weeks now. In this short time frame, I created a bunch of silly bots to test what GPT-3 is capable of.

I've made Epic Rap Battle bots, charade bots, cover letter generator bots, Processing code generation bots, marketing bots and capitalization bots.

I've also been able to condense some of my thoughts about what I think GPT-3 truly represents for the future of humanity, and our interaction with computers.

The Real Magic of GPT-3

I believe that the real use case for GPT-3 going forward won’t be vertical solutions to a business. As of July 2020, GPT-3 is too unreliable to be trusted with major tasks like running an organization or managing a large team. GPT-3 won't be replacing entire software engineering or design teams any time soon either.

I think the true power of GPT-3 lies in its ability to create atomically small bots to answer problems that people face in their daily lives.

My most recent experiment with GPT-3 was a collaboration between RunwayML and GPT-3 to create visual poetry. While I was switching tabs between RunwayML and the OpenAI API, I felt like I was watching different band members perform their specific instruments at a master level. I was the maestro, coordinating the timing of each player, and coaxing out their best performance.

a visual poetry piece I made using GPT-3 and RunwayML

Each model had its narrow use case. One generated portraits. Another turned faces into flowers. Another yet wrote poetry in the style of Rupi Kaur. On their own, all these models produced cute results that are fun to look at, but quite shallow.

Together, they became much greater than the sum of their parts, and began to tell a story.

The Landscape of Personal Assistants

As of July 2020, there are three prevailing "personal AI assistants" on the market: Siri, OK Google, and Alexa.

The problem with Siri and OK Google is that they are much too general. You never know what you’re going to get back. Most of the time, Siri throws its metaphorical hands up in the air and searches the web with your exact search string. At other times, it extracts keywords from what you said and returns an info card with an answer. Sometimes, it misunderstands your point all together and you get results like this:

The trouble with Alexa is that you have to go and manually find apps that solve your particular problem set. If you want to make your own, you’ll need a working knowledge of AWS and Machine Learning. If you're not interested in learning how to program to create an Alexa skill, you're kinda SoL.

GPT-3 has solved this space.

It uses real context clues from conversation, rather than just giving up when it can't find a predefined response for a set of keywords. It is flexible, and can be for as specific a use case as you want it to be. It's easy to pick up a conversation with it, anytime and anywhere.

Most importantly, it puts the power of creation in the hands of those facing the problem.

Power to The People

Everyone faces different problems.

“I lived in a mile-long village in the middle of a western province in Kyrgyzstan: there were larch trees on the snowy mountains, flocks of sheep crossing dusty roads, but there was no running water, no grocery store. The resourceful villagers preserved peppers and tomatoes, stockpiled apples and onions, but it was so difficult to get fresh produce otherwise that I regularly fantasized about spinach and oranges, and would spend entire weekends trying to obtain them.” - Trick Mirror by Jia Tolentino

The GPT-3 bots a citizen of Kyrgyzstan might find useful are much different than the bots a 26 year old creative coder in the US would.

And yet, software engineers on small teams across the world create apps and solutions that billions of people use, and they just assume that it works for everyone.

Since problems can’t truly be anticipated, it’s best to give people the power to create their own solutions.

This is the real power of GPT-3. It is easy to use. Anyone with a foundation in human language can communicate their desires to it. With a little bit of practice, anyone can learn how to split their problem into its component parts, and then feed GPT-3 a few examples of that problem and a solution.

This lowers the bar for people significantly. If you can define a problem, and a suitable solution, GPT-3 can solve it.

What may this look like in practice? I'm glad you asked, my friend. Here's what I'm thinking...

My Own Concierge

This is my concierge. It's for me. Not for my cousins. Not for my neighbor's dog. It's for me. It solves problems that I want solved, and helps me with things that I want answered.

In a nutshell, each page runs its own seeded model. These models are wildly specialized, that’s intentional.

And the best part of all? Removing and adding modules is as simple as setting new goals, and asking different questions.


GPT-3 is a game changer. The people on Twitter are right. But a lot of the tweets I’ve been seeing involve vertical solutions. I’m suggesting something much less ambitious, but I also believe much more realistic and powerful.

By giving individuals the power to solve their own problems, GPT-3 makes everyone the owner of their own little universe.

One day we might all have our own digital concierges.

These concierges will grow and change with our problems and interests, following us from primary school mathematics to who to take and what to wear to prom to what to do with our 401k allotments after retirement.

To me, GPT-3 represents a potential lifelong partner created by you, for you.

And that’s pretty sweet.