It’s finally here! I recently published my poetry book Poetic Genesis: An Anthology of 52 AI Crafted Poems on Amazon. Generating all the poems took approximately two days. It was the ideation, curation and publication stages that actually took much longer. However, apart from the human time and energy cost, I’m glad to say $0 was spent to make this anthology. So how did I do it? Read on to find out how I made a poetry anthology using ChatGPT!
The origin of Poetic Genesis
Like many who keep up with tech, I came across ChatGPT (a powerful text generation chatbot) early last year on Twitter. It was the coolest application of AI that I’d ever seen! I immediately started playing with it and it truly felt like magic to me. From time to time, I’d also check out what others were making on social media and the creative expression on display was astonishing. One such tweet that stood out to me was Andrej Karpathy’s tweet which showed that ChatGPT could actually write poetry!
I was amazed. This example literally gave me life! Why? Well, back in my high school days, English language and literature were my favourite subjects. Although I later gravitated towards maths and sciences, I still had a soft spot for linguistic expression. Furthermore, my interest in poetry had recently been strengthened during the pandemic. In the darkest days of this unprecedented disaster, I’d started writing poetry to help me grapple with uncertainty and give concrete form to my emotional experience.
So when I read Karpathy’s AI-generated poem, my brain was primed. My curiosity skyrocketed to max level! I became a lot more intentional about my ChatGPT exploration. I wanted to know everything about its poetic capabilities. How many different poetic forms could it produce? Was its knowledge base comprehensive enough for compelling poetry? I was hooked. I felt alive like never before. The possibilities seemed endless. I kept experimenting and learning. And lady luck blessed me again. I stumbled upon Andrei Lyskov’s epic tweet. This thread described how he and Lucas Cohen made a whole book of aphorisms in 12 days using ChatGPT and Stable Diffusion (an open-source image generation tool). Their book is called The AI Sages: Eternal Truths for Modern Times.
This tweet thread answered a question I’d been mulling in my mind as I played with poetic prompts. How could I get better with ChatGPT and simultaneously make something of value? Andrei’s tweet clarified a lot for me. I had received another jigsaw piece but the puzzle was still incomplete.
Learning from my initial poetic experiments
My initial approach when trying to create poems was to prompt ChatGPT for the style of a specific poet e.g. Shakespeare, Rumi, etc. This turned out to be problematic…even when I looked at other celebrated poets in the public domain (like those listed below) the results weren’t particularly striking. I hypothesise that this is due to ChatGPT not being fine-tuned with enough training examples from these poets:
This was my first wall. My curiosity was still high and I remained adamant in my experimentation. I changed my approach. Instead of prompting ChatGPT for a specific poet’s style, I asked for various poetic forms e.g. sonnets, haikus, etc. These results were significantly more engaging. Unfortunately, the variety still wasn’t enough. As I soon found out, there’s a limit to how well ChatGPT can produce poems of certain forms. This meant I was limited to about seven poetic forms. This wasn’t enough variation, I needed more. So I went looking for innovation elsewhere. It wasn’t long before I realised the simplest way to solve this design challenge was to grow the generative universe by covering multiple topics.
Generating engaging poetic content
With any product, the most important thing to consider is value. And value, like beauty, is in the eyes of the beholder. So I had some questions to answer: What topics should I choose to engage readers? How was I going to choose these topics? Who would want to read the topics I ended up choosing? As I pondered the possible answers, I suddenly recalled something useful I’d stumbled upon during a past internet escapade; it was a neat tool by Google. This tool was Google Trends – an interactive dashboard – which enables anyone to see the most popular terms that have been searched for across the world. It also allows you to examine this data across different periods with handy visualisations and lists. Furthermore, Google publishes an annual Year In Search breakdown, like this one for 2022 across categories. I intuitively knew adopting this tool was the right call. Why? It was a unique solution to solving my topic problem. It clearly aligned with people’s proven interests. This data-driven mirror reflected the consciousness of digital humanity. Well, at least those who used google search. It would be my topic oracle and with that, another jigsaw piece fell into place.
A few days after seeing The AI Sages book and deciding on my topic-based approach, the rest of the pieces arrived, one after another. A proverbial lightning bolt of inspiration struck my consciousness with the power of a supernova! Through continued refinement during the project, these divine ideas later evolved into the following execution strategy:
- Source potential topics for poems from the global Google Trends: Year In Search e.g. Harlem Shake
- Use ChatGPT’s generative power to make a poem on this topic in a given form e.g. a haiku
- Repeat steps 1) and 2) until I have a selection of poems which are interesting across different poetic forms
- Examine the poems generated and make new ones if need be until I have a collection that makes sense holistically
- Utilise the free trial from the canva graphic design suite to put the poems into a draft manuscript
- Get initial feedback on the draft manuscript from my friends and network
- Keep iteratively refining the manuscript based on feedback
- Design the cover using Playground AI’s generative image tool
- Use Amazon’s Kindle Direct Publishing tools to publish the world’s first ChatGPT poetry anthology!
- Share my creation with the world on my platforms and beyond 🎉
And so my quest began. A quick internet search confirmed that AI poetry anthologies already existed. A Korean firm had already made one. On Amazon, author Shane Neeley had already published his own anthology. This was fantastic. There was precedent for my work. But my anthology would be different. Whereas the Korean firm and Shane’s collections involved multiple human collaborators, my poetry would break into uncharted territory. I would primarily leverage ChatGPT. This had never been done before! I remember being in a sort-of myopic rapture as I envisioned how the future of poetic literature was about to change. I had to strike while the iron was hot. I had to be the first to launch before anyone else!
On a slightly personal note, I needed this project. I was going through an extremely challenging period of my life. Poetic Genesis became a sacred fountain that fed my spirit during the external trials and tribulations I faced. Once again, poetry became a key anchor for me but this time was different. This time, there was a cool technological flavour. Making Poetic Genesis kept me grounded and was a healthy conduit for my creative energy. Throughout the weeks that followed the birth of this idea, there were many times when I felt like giving up. Sometimes factors outside my control would shake my confidence. A few times I even deliberated about making a mediocre product. However, my standards and vision were too strong. I had come too far. I held the line.
At other times, it felt as if the universe was guiding me. As the small successes grew, I realised I had no choice but to complete my mission. I would make the highest calibre anthology I could despite the resource constraints and setbacks. My circumstances no longer mattered. Only my will to overcome obstacles mattered. Only my drive to see my vision through to completion mattered. I would find a way no matter what happened. And so I carried on.
Developing my prompt engineering
After many sleep-deprived nights, I got used to executing the strategy from steps 1) to 5) and learned a lot about prompt engineering to coax the results I wanted from ChatGPT. I even figured out some prompt techniques to generate poems on controversial figures. For example, I was able to generate poems on Gaddafi such as the one below. In the final book, I elected to avoid such poems to make the first version as accessible as possible.
However, running into such blocks on certain topics was instructional. It helped me to understand some of the perspectives and value systems in-baked into ChatGPT. This gave me insight into the limitations and safety guardrails within the system. I was glad that OpenAI had still shipped despite the reputation risk. I was having a lot of fun. Other more benign requests for topics like the popular Squid Game TV show presented a slightly different challenge. Why was this? It turns out the research preview had only been trained with data up until 2021. So for 2022 trends like Squid Game which were absent in the training data,, there was no reference to assist ChatGPT to output a relevant poem. In this case, prompt engineering required knowledge injection. This imbued the session with domain knowledge to facilitate smoother poetry generation. An example prompt is given below:
Bringing all the elements together
All the poems were generated. It was now time to get feedback from my friends and Asiatu Koroma (my editor). This would be essential for finalising the first version of the manuscript. This first version included both the front matter and back matter which you’d normally find in a book. With the great feedback I received, I was able to make key decisions on factors such as the order of poems, the book’s overall structure and the value proposition for potential customers. A few more manuscript revisions later (approximately eight revisions) and the core manuscript reached release quality.
I then took the value proposition ideas from earlier discussions and fashioned them into a product description. It was also around this period that I decided on the anthology title. I chose Poetic Genesis. Why? Because I wanted this work to represent the next frontier of poetic creation! On the other hand, I wanted the subtitle to hint at the novel way this anthology was created.
The only thing left to do now was to design the cover. I’d already been experimenting with AI-generated art. I’d experimented with Dall·E 2, Dreamstudio and was now using Playground AI. However, I wasn’t sure where to start. How could I make an appealing cover? The type of cover that would stop someone from skipping past my work and instead encourage them to have a look. Thankfully Asiatu came to the rescue. She provided a really cool seed prompt which I expanded upon. After several iterations, comparisons and discussions, the cover image was finally done.
Every necessary element was now ready for the Kindle Direct Publishing (KDP) program. This program allows anyone who wants to self-publish on Amazon to edit book details, organise book content and set up pricing across Amazon e-commerce stores. It also comes with several handy tools which make book creation much easier. I used Kindle Create to compile my book content and output a publishable file which I then uploaded to Amazon. Understanding the intricacies of the KDP program was quite challenging but fortunately, Paul Marles’ excellent youtube channel – especially his book publishing tutorials playlist – helped me to navigate the complexity. With all required details set and files uploaded, Poetic Genesis was now ready for review by the Amazon team. This took approximately 36 hours and then the long-awaited moment arrived. Poetic Genesis: An Anthology of 52 AI Crafted Poems was live. I could hardly believe it. I’d taken the seed of an idea, watered it with my life-force energy and now a young sapling had sprouted into reality. I was beyond ecstatic.
Sampling from a ground-breaking anthology
Poetic Genesis features poetic forms from limericks to haikus, from ballads to odes and more! Covering legends from Kobe Bryant to household names like the PlayStation 4, there’s a poem for everyone in this nostalgic anthology. I am yet to see anything like it among the world’s current poetry anthologies. Which other work has a poem on El Chapo or the fidget spinner? As I know we all love teasers, here are a few to whet your sophisticated appetite. We’ll start with the Harlem Shake 🤣 In case you don’t remember, I’ve included a video below to jog your memory of this sensational meme that took the world by storm.
What a throwback, right?! The Harlem Shake is still lit to this day but although the original video came out in 2010, it wasn’t until 2013 that it really took off and made it to spot #5 on the global top trends! Of course, I couldn’t resist asking ChatGPT to make a poem on this classic and here’s what it came up with:
For the last teaser, I’ve included a topical city which captured our hearts in 2015; the city of Paris. Why did this hold our imagination? Sadly, this was due to the terrorist attacks that occurred across the city in quick succession. However, instead of dwelling on the darkness, I wanted to highlight the magnificence of Paris. I’ve found that appreciation especially in tough times is key to growing, sharing and protecting the light in this world. I hope you enjoy it:
The present-future of poetry
As the poetic samples above show, AI-crafted poetry is here and it’s improving rapidly. We stand at the cusp of a seismic shift in how humans everywhere create and consume. Making Poetic Genesis has shown me that poetry can be even more accessible than it is currently. These generative AI systems can help us all explore the space of ideas in ways we’ve never dreamed of before. I believe this is truly valuable.
In our world, there are relatively few people who write poetry. This means most people are missing out on a powerful way of creative expression. The majority of concepts are never poetically explored because there’s no one even looking at them. Is there any reason why the things you care about can’t be expressed through poetry? There’s no good reason that comes to my mind. Is there any reason why customised poetic systems tailored to your preferences can’t be a part of your day-to-day life? I don’t think so.
In an age of fast consumption encouraged by social media platforms like Tik Tok, custom poetry can help us slow down. Breathe. Listen and contemplate. It could also help us face our inner demons. Grant us the power to confront issues we’d usually never examine. Equip health practitioners with a personalised effective tool to integrate into a more holistic patient-centred healing system. Writing is both a means of expression and comprehension. In the form of poetry, it can also heal, inspire and connect. This is just the start. With our imagination, I’m sure we’ll conceive of more ways to make our experience richer with AI-crafted poetry.
If you’re still not convinced or want to know more, check out Poetic Genesis on Amazon. Explore the cutting edge of human-machine collaboration and get a sneak peek into the future of literature. You’ll be pleasantly surprised. This anthology will take you through a whirlwind tour of the past 13 years of trending search topics. Poetic Genesis will have you elated and at times, wistful from this nostalgic exploration of our shared past. It’s the perfect accompaniment for everyone who enjoys poetry and for anyone who likes technology. It’s sure to become a treasured addition to any literature lover’s library and help you to see artificial intelligence and our world in a different light. You can purchase your copy here.
I’m always up for speaking and connecting with others interested in the possibilities of AI. You can find out more about me here. Feel free to reach out to me on twitter or via my other links. Thank you for reading.
Pingback: Exploring foundation models symposium: Research insights & wonderful conversations - Azuremis