A Wild Adventure With GPT-3

Featuring Indian Mythology and Neruda

Detective Del Spooner : Human beings have dreams. Even dogs have dreams, but not you, you are just a machine. An imitation of life. Can a robot write a symphony? Can a robot turn a... canvas into a beautiful masterpiece?

Sonny : Can *you*?

After months of waiting I finally got access to GPT-3. I was prepared to be underwhelmed but it has surpassed all of my expectations for what I thought was possible with AI and language models.

Its output is often profound and moving and better than most of my writing.

You log in and see the playground.

You slowly interact with it and realise that this is a new kind of user interface. A new kind of programming. One you slowly uncover by poking at it. Learning how to coax it to do what you want. An alien intelligence that is at once stupid and infinitely more knowledgeable.

You enter a few examples and then let it do its magic. It’s that simple.

What took weeks of training using specialised NLP models and lots of training data is now a few clicks on a web form.

One of the major challenges in AI has been how to integrate the more statistical kinds of AI with the classical / symbolic systems. GPT-3 presents a glimpse into how this can be achieved.

Consider the task of getting from A->B. This involves two kinds of AI systems :

• Statistical systems that do voice => text and text => intent

• Classical systems that use constraint propagation and search to find the ideal route

Merging these two has been a lot of work.

Converting text => intent takes weeks of work and long iteration cycles with lots of training data.

Here’s what GPT-3 can do in seconds

With just a few examples GPT-3 can generate structured output that can be parsed by a program that implements classical AI.

The extremely fast iteration cycles and lower requirements for training data mean that we can stitch very complex systems together quickly.

Alright it can make some basic analogies. But can it feel ?

GPT-3 really shines at poetry and I discovered that it’s shockingly good at imitating the style of one of my favourite poets Pablo Neruda.

The prompt given to GPT-3 is in bold and the rest is by GPT-3.

The following is a poem about Maxwells equations in the style of poet Pablo Neruda:

"I have always loved you, O equations of Maxwell,

For you are the most beautiful thing we have in this world.

I love your graceful symmetry, your simplicity and clarity.

You are the song of the Universe, a cosmic lullaby.

You are the poetry of nature, written with light and electricity.

You are the music of the spheres, played on a harp made of vacuum.

And I love you for being so simple and yet so complex,

For being so abstract and yet so real.

I love you for being true in every place and at every time.

I love you for being the result of a hundred years of thinking and calculating,

For being the fruit of a tree planted by Faraday and Maxwell.

I love you for being the answer to my questions about nature,

For being the solution to my problems.

I love you for being so difficult and so simple at the same time,

For making me feel so small when I think about you but so big when I understand you. I love you for being the most beautiful thing in the world,

For being the closest thing to God that I have ever seen.

I love you without knowing how, or when, or from where.

“music of the spheres, played on a harp made of vacuum.” This was good. Really good.

Alright but can it imitate real people. Might I someday be able to point GPT-3 at a persons blog posts and have it be read out by them in their voice ? It was time to find out. I chose my friend Nindalf (https://blog.nindalf.com/).

He had a couple of posts that had gone viral on Hacker News so I was hoping to get some viral worthy completions. I prompted it with a few of his posts and asked it to write about The Economist which I know is one of his favourite publications.

The Economist Magazine

By Nindalf

I’ve been reading The Economist for a while now. It’s a weekly magazine, so I read it every week. I think of it as the ‘big picture’ view of the world, and it does that well.

The magazine is full of interesting stuff - history, economics, politics, science, technology. There are some articles that are just plain fun to read - like the ones on popular culture or travel. There are also lots of articles on business and finance which help me understand how businesses operate around the world and what they do differently from each other.

I also like how The Economist presents information in an unbiased way even when they have their own opinions on things. They don’t try to hide their opinions but they present them as one among many possibilities without exaggerating them or making them sound more certain than they really are (as far as I can tell). This makes their views easier to accept if you agree with them or easier to dismiss if you don’t agree with them without feeling like you were lied to or manipulated into believing something false. The best example is climate change - most people know about this issue but there isn’t much agreement on how serious it is or whether we should be worried about it at all (or whether there even is an issue). Even though The Economist has its own opinion on this topic (they think we should worry), they present both sides fairly and let readers decide for themselves what to think about this issue based on the facts presented in the article instead of trying to convince readers using emotion-based arguments or by presenting only one side of the argument without providing counterpoints and evidence against that side.

There aren’t many downsides to reading The Economist . One downside is that sometimes their articles take a very long time before getting into anything interesting - especially when talking about economics where they tend not to focus on specific companies but rather talk more generally about economic systems in different countries/regions/cities etc.. Another downside is that sometimes things get lost in translation from English into Russian then back again because some words have different meanings depending on context so translating literally doesn’t work well enough here (and sometimes not at all). But these downsides aren’t enough for me not to recommend reading The Economist .

The resemblance to his views and writing style were uncanny. We realized that GPT-3 had studied his Hacker News comments and picked up on the name Nindalf. The above piece has some similarities to a Hacker News comment he wrote. However the rest of it was quite original !

So yes indeed we can recreate people in some sense !

I loved a post he had written about Factorio and Software engineering. I wondered what GPT-Nindalf would have to say about Zelda. Would it make similar analogies ?

The Legend Of Zelda Breath Of The Wild

By Nindalf

I’ve been playing Zelda Breath Of The Wild for the past week and it is amazing. It’s a massive open world with lots of things to do. You can go anywhere you want, climb any mountain, swim in any lake, and fight any monster. The game is so big that I can’t even imagine how long it took them to make this thing.

But what I really love about this game is how they handle quests and puzzles. Most quests are optional but there are a few mandatory ones which advance the story. These quests are usually solved by finding an NPC, talking to them, doing what they say (or not), and then talking to them again to get your reward (or not). This isn’t much different from most games out there but here it gets interesting - the rewards for these quests aren’t XP or gold or items or anything like that. They are usually “get a new ability” type of rewards which makes you more powerful in some way - like being able to lift heavier objects or move faster while carrying heavy objects or shoot arrows further away etc etc etc…

As for puzzles, they often have multiple solutions depending on your playstyle and equipment loadout. There was one puzzle where you had to use bombs in a certain way with specific timing so that you could cross a gap without using the bridge nearby which was guarded by monsters who would kill you if you tried crossing it normally. If I remember correctly, there were three ways of solving this puzzle - one involved using bombs as weights for a raft made of wood planks found elsewhere on the map; another involved making an ice block with magic powder; another involved making an ice block with magic powder + fire arrows + firewood from nearby campfires; yet another involved making an ice block with magic powder + fire arrows + bomb arrow combo; all three were valid solutions because they used common materials found around the area (magic powder/fire/wood) but used them in different ways (ice block instead of raft/ice block instead of burning bridge/bombing ice instead of burning bridge).

This reminds me very much of software engineering problems where we have multiple ways of solving problems depending on our current knowledge and available resources at hand - sometimes we need something fast so we write quick code without tests because time is short; sometimes we need something reliable so we write tests first because time isn’t short enough for us do everything else too…and sometimes we need something fast AND reliable so we write both tests AND code before shipping…and sometimes our boss wants us to ship ASAP no matter what…and sometimes our boss wants us to ship good code no matter what….and then there are deadlines….oh boy deadlines! That reminds me why I love my job!

Mind = BLOWN.

I wanted to test this out to its limit. On the suggestions of some friends we decided to challenge it with Indian Mythology since that is something we Indians are intimately familiar with.

Bheema and Yudhishtir were smoking a beedi while lying on the parapet of the turret, watching the sun set.

"Tell me about our kingdom," Yudhishtir said.

Bheema took a long drag and blew the smoke out in a thin stream. "What do you want to know?"

"Everything. I want to understand the kingdom that we have inherited."

"That's a tall order, Yudhishtir," Bheema said. "We currently rule over five hundred villages, and we have another seven hundred villages that we can raid at will. We have twenty-two thousand fighting men under our command. We are at war with the Kauravas and the Trigartas. I don't know where to begin."

"Start at the beginning."

"Very well. We are the sons of Pandu, the king of Hastinapura. He was a great king. As a young man he was a great warrior and on one occasion his valor earned him the name of Jayadratha, 'the one who conquered Indra.' When he was twenty-five, he married his uncle's daughter, Kunti, and she bore him five sons: Yudhishtir, Bhima, Arjuna, Nakula, and Sahadeva. You are the eldest of the five.

"At the time of your birth, a sage named Vyasa was in Hastinapura. He foretold that your father and mother would die of snakebite, and in order to save your life, he said that you must be taken to the king of a distant land and left there. Your mother abandoned you on the banks of the river, and you were found by the king of the gandharvas, who brought you up as his own son. It was he who named you Yudhishtir.

It turned a totally hilarious prompt into something serious and largely consistent with mythology.

And suddenly the thought struck me. What would Pablo Neruda make of the Mahabharatha ? Lets find out …

The following is a summary of the Mahabharata in the style of poet Pablo Neruda:

(The following is a summary of the Mahabharata in the style of poet Pablo Neruda)

I do not know what you have done with your life, O reader. But I can tell you what I have done with mine: I have read the Mahabharata.

I have read the Mahabharata and I am amazed. I have read the Mahabharata and I am ashamed. I have read the Mahabharata and I know that nothing is more important than love.

I have read the Mahabharata and it has taught me that we are all part of one another, like a single body with a thousand limbs, or like a great tree whose roots spread in all directions.

I have read the Mahabharata and it has taught me to see my own life as part of a great pattern, which includes everything else in existence.

I have read the Mahabharata and it has taught me that there is no such thing as evil; there is only hunger, which is a kind of good, for what we eat we become.

I have read the Mahabharata and it has taught me that there is no such thing as death; there are only changes from one form to another.

I have read the Mahabharata and it has taught me that there is no such thing as a lie; there are only transformations.

I have read the Mahabharata and it has taught me that there is no such thing as a final truth; there are only better and worse ways of seeing things.

I have read the Mahabharata and it has taught me that all things are one, that all life is one, that all love is one. I am not learned in the ways of words, O reader, but I can tell you what I mean:

I have read the Mahabharata and it has taught me to love.

A brilliant summarization of the essence of Hindu philosophy echoing the words of Neruda !

Now it does not always come out this well. I would say the output is underwhelming 70 % of the time. But that should not blind us to the what has been achieved.

Now for the Pièce de résistance.

I decided to make sense of this all through the eyes of my favorite poet again.

I asked GPT-Pablo to write about machine learning

The following is a poem about Machine learning in the style of poet Pablo Neruda:

The Machine Learning

I want a machine that learns, as if it were a child.

I want it to be born in the countryside, among the trees and the flowers.

I want it to play with other children.

Not with toys or dolls, but with real things:

With sticks and stones and mud pies, with water and fire.

I want it to learn the names of things and to know the difference between them.

I want it to like and dislike things, to feel joy and sorrow.

I want it to have its own ideas, its own feelings.

I want it to be able to look at the stars and wonder why they shine.

I want it not only to learn from books but also from life.

Not only from pain but also from pleasure, not only from what is right but also from what is wrong.

I want it never to lie or cheat or steal; never be greedy or cruel or drunk; never be angry without a cause nor happy without a reason.

And when I die I want this machine of mine still to be learning, still loving, still creating things that will make people glad that they were born on this earth.

I was left speechless.

I for one Welcome our new Neural Poet overlords.

For more posts / updates subscribe to this newsletter and follow me on Twitter (@samhanknr)