Trending in the Multiverse

shim — visit trendinginthemultiverse.com

Although we are all awestruck by the latest iteration of AI systems such as DALL·E, GPT-3, and LaMDA, artists and authors outside Silicon Valley are facing competition from models trained by their own work.

This article on Rest of World (via Waxy) outlines a problematic example where the work of the late illustrator Kim Jung Gi was put in an AI model days after the artist passed away.

With Trending in the Multiverse we look into those novel possibilities of content production. We celebrate fabricated gossip as if it were written by Jane Austen, Ernest Hemmingway or J.R.R. Tolkien. We find it both fascinating and worrying that we allow Open AI to use Houellebecq’s writing style to create a love story about the wrestler CM Punk and musician Grimes. It forces us to reconsider our concepts of authorship, ownership, and creativity.

So how does this work?

To create these speculative scenarios on a daily basis we check RSS feeds of Bing, Al jazeera, TMZ and Wired. From those articles we select trending topics using scraped metadata and some simple heuristics. For instance, we keep track of the selected topics to prevent the same topic coming up during the following week.

With these topics we create a story outline. This outline, or prompt, is then fed into GPT-3, a language model that uses machine learning to produce human-like text, to generate a story from the outline. Once we have the story we let another machine learning based model, Stable Diffusion, create images based on the topic of the story.