Welcome back to the /promptcollective newsletter! For our regulars, thanks for sticking with us. If you're new here, we're thrilled to have you. At its core, the /promptcollective newsletter delves into human-AI interactions and collaborations within the creative industries.
New opportunities = new challenges in literature
Today, instead of an experiment, we're presenting a "think piece" on literature in the age of AI. With Large Language Models (LLMs) in our midst for just under a year, it's intriguing to see the focus on literature.
This article aims to explore a few concrete examples and debates that I have stumbled upon, surrounding AI in literature and literature in AI. These by no means cover the entire area, but they do provide different perspectives on the subject matter at hand.
One of the examples has left the Swedish literary society in a real state this week, as Camilla Läckberg, the reigning queen of Swedish crime fiction, has been put on the spot. And it started with an analysis of her work - performed by an AI…
By the end of the newsletter, I will offer a few suggestions for further progression in the field.
AI-Generated Publications
To call the current LLMs merely "text generators" would be a simplification. Yet, most users encounter them primarily through their text outputs. This has inspired many to create literary works using LLMs.
In December 2022, Ammaar Reshi released the children's book "Alice and Sparkle," crafted over a weekend with Chat GPT and Midjourney. The ensuing debate, fueled by Reshi's announcement tweet, centered on the data sources used to train LLMs. Illustrators voiced concerns over the unauthorized use of their creations to train the LLMs used.
In spring '23, Swedish publisher Novellix launched four short stories inspired by Swedish literature classics, all penned by GPT-4. However, these stories were criticized for their subpar quality, suggesting that AI might need more human guidance for literary tasks.
Concerns Over Data Theft
A contentious issue is the material used to train LLMs. It was disclosed that one dataset for training the LLM "Llama" included pirated library Books3. Alex Reisner of The Atlantic highlighted this with his apt remark: "The future promised by AI is written with stolen words." This is a real problem and even though there have been attempts at removing the illegal Books3 library, it seems the current version of “Llama” still carries the library as part of its training data.
The Advent of AI Audio
Since 1971, Project Gutenberg has offered e-book versions of public domain literature for free. This summer, in collaboration with Microsoft and MIT, they launched free audiobooks using advanced neural text-to-speech technology. With AI narrating, they released 5,000 audiobooks – a feat previously unachievable due to costs and time constraints. While this initiative democratizes access to audiobooks, it has also sparked debates about the potential loss of jobs for voice actors, right in the middle of the Hollywood actors being on strike.
The AI-generated voice isn’t as perfect as a human voice, but why don’t you judge for yourself and take a listen to the AI voice version of the D.H. Lawrence collection of short stories “England, my England”? Once you’ve finished this newsletter of course.
Originality and Ghostwriting
There's a romantic notion that literature should be original, born from a single creator. Recently, in Swedish literary circles, an AI analysis suggested that best-selling author Camilla Läckberg might have used ghostwriters for some of her works. Interestingly, the music industry has long accepted ghostwriters without any stigma.
Sidebar, (mainly to the hip hop heads out there) did you know that Dr. Dre’s classic “Still D.R.E.” was penned by Jay-Z?
What's Next?
The intersections of AI and literature are multifaceted and evolving. Transparency regarding training data for LLMs is crucial, both for IP rights holders and to foster trust. Distinguishing between plagiaristic AI outputs and genuine inspirations will be challenging, yet essential.
The very terminology we use, like "ghost in the machine" or "magic," needs reevaluation. Literary works, whether human or AI-assisted, should be valued for their content. Initial forays into AI-generated literature highlight the indispensable role of human creativity. As suggested in a previous edition of this newsletter, proper labeling of literature sources might be an avenue worth exploring for tech companies, publishers, and authors alike. That way speculation and disbelief could get suspended and the current and new AI tools could be treated exactly as new tools rather than something more dubious.
The Project Gutenberg initiative, converting titles into audiobooks using AI, exemplifies the positive potential of such collaborations, broadening access to literature. And doing it with incredible speed.
Now, that was a whole lot. Now to…
A Lighter Note
The AI Reddit community was buzzing this week, with a post titled "Is this AI or not?" garnering over 2,700 comments. Dive into the debate and decide for yourself!
Thank you for joining us on this journey!
Please subscribe if you haven’t done so already.
Who Are We?
The /promptcollective was founded by Jes Brandhøj (Denmark) and Hannes Jakobsen (Germany). We're on the lookout for like-minded enthusiasts. If you're passionate about the AI-creative nexus, reach out at join@promptcollective.xyz.
Ps. On the topic of labeling: This newsletter was written by a human, using GPT-4 to proofread, correct grammar, and suggest improvements. Image number one was created using Midjourney.