The Foundation

OpenAI, for those unaware, is an Artificial Intelligence (AI) research foundation founded by Elon Musk, Sam Altman, and others in 2015. The foundation conducts research in the field of AI with the stated goal of promoting and developing friendly AI in a way that benefits humanity as a whole.

Over the weekend, I was able to play around with two of OpenAI’s flagship products, GPT-3, and DALL-E.

GPT-3

GPT-3 (or more formally, Generative Pre-trained Transformer 3) is OpenAI’s third-generation language prediction model in the GPT series.

Loosely speaking, you can give GPT-3 an input such as “Write a description for an introduction to programming class” and it will generate it with pretty high confidence.

Using the OpenAI Playground, I prompted it to create some short stories and it did surprisingly well.

Results

Here’s what I asked, and how it handled it.

Prompt: Write a synopsis for a short story about a honeycrisp apple’s first day of school.

When Honeycrisp Apple is sent to school for the first time, she is excited to learn and make new friends. However, she soon realizes that the other apples in her class are different from her. They are all red apples, and Honeycrisp is the only green apple. The other apples make fun of her and call her names. Honeycrisp is sad and feels like she doesn’t belong. But then, she meets a boy named Red Delicious who is also different from the other apples. He is red, but he doesn’t make fun of her for being green. Red Delicious and Honeycrisp become friends, and she finally feels like she belongs.

(Disney or Pixar — if you steal this idea at least Buy Me a Coffee)

Prompt: Write a short story about a cat surfing on an asteroid in space.

The asteroid belt is a lonely place, but it’s home to one intrepid feline. This kitty loves nothing more than strapping on a space suit and hitting the waves – of space dust, that is. With a little help from her rocket pack, she can surf the asteroids for hours, enjoying the solitude and the view.

But one day, she gets a little too close to a large chunk of rock and is pulled in by its gravity. She’s now orbiting the asteroid, and her only hope is to use her rocket pack to break free. With a few well-timed bursts, she finally breaks free and continues on her merry way.

Wow! I’m both amazed and a little saddened by how creative GPT-3 is. I’m not sure I would have been able to come up with a short story like this.

I was curious how GPT-3 would handle songwriting, so I asked it to write a pretty generic song about driving around at night in the summer.

Prompt: Write a song about driving in a convertible with the top down on a summer night.

Can’t you feel the wind in your hair?
The moonlight shining on your face
The stars up above
Just shining down on us

We’re driving in a convertible
With the top down
On a summer night

We don’t care where we’re going
We’re just enjoying the ride

The wind is blowing through our hair
The music’s playing loud

We’re laughing and singing
All we need is this moment

We’re driving in a convertible
With the top down
On a summer night

I think it’s safe to say that GPT-3 is an amazingly powerful (and a bit scary) tool.

Next, I got to look at a mini version of DALL-E.

DALL-E

Recently, the OpenAI foundation released DALL-E 2. DALL-E 2 is the upgraded version of DALL-E, which is a 12-billion parameter version of GPT-3 that can be used to generate images from a textual description.

Unfortunately, I’m still on the waitlist for access to DALL-E 2, but I was able to take the first version of DALL-E for a test run.

I was able to generate images using the DALL-E Mini, which is a smaller version of DALL-E Mega (DALL-E mega can generate much higher fidelity images, but requires way more computing power) so my results were mixed. You can read more about the DALL-E playground used to generate these images on its GitHub Repo page.

Results

Here’s what I asked it to create.

Prompt: An illustration of a cat

Prompt: A honeycrisp apple

Prompt: An illustration of a sloth snowboarding down a mountain

Prompt: A sunset on the beach

Prompt: A bust of Homer

Prompt: A realistic Homer Simpson

And that’s enough nightmare fuel for the day. As you can see, the results are pretty mixed. The results ultimately depend on the quality of the prompt and the data used to train DALL-E.

You can read more about DALL-E and its (amazing) capabilities on its website.

The future of OpenAI’s products looks both promising and scary, and it will be very interesting to see how it impacts our lives.