If you've seen a lot of fancy images these days that look like they were drawn by a little kid, know this: what you've seen is the latest demonstration of the capabilities of artificial intelligence. Capabilities, in this case at the service of trolling.
Hugging Face is the company behind this modern tool: DALL-E. This company hosts open source projects in the field of artificial intelligence, and among their creations, they saw a tool that exploded the traffic on their server.
This deceptively simple program generates nine images in response to any typed text, and although it is now widely used, it was started almost a year ago by an independent developer.
DALL-E, meme creator
But after recent improvements and some viral tweets, the tool has become the biggest source of funny and weird memes on the internet. No one is a stranger to DALL-E, the artificial pictorial intelligence.
As more and more people created and shared DALL-E Mini photos on Twitter and Reddit, and new users joined in, Hugging Face was overwhelmed by the traffic on its servers.
"Our engineers were up the first night," Clement Delang, CEO of Hugging Face, told Wired. In recent weeks, the DALL-E Mini has generated about 50,000 images per day.
What does the future hold for DALL-E?
DALL-E Mini's viral moment doesn't just herald a new way to create memes. It's also a preview of what could happen when AI tools that create images on demand go mainstream, and a reminder of the uncertainty of their potential impact.
Algorithms that create personalized photos and artwork can transform art and help businesses with marketing, but they can also be manipulative and deceptive.
The beginning of mass manipulation?
A warning on the DALL-E Mini website states that a film can "reinforce or exacerbate social prejudices"or "create images that stereotype minority groups".
DALL-E Mini is inspired by a more powerful AI imaging tool called DALL-E (an alliance between Salvador Dali and WALL-E, they were good friends), which was introduced by the research company OpenAI in January 2021.
Although OpenAI has not released any code, it successfully built the first version of DALL-E Mini at a hackathon hosted by Hugging Face and Google in July 2021.
Images still... blurry
The images of DALL-E Mini clearly look disconcerting. Objects are often distorted and blurry, and people appear with missing or damaged faces or body parts.
But it's still possible to recognize what's being depicted, which is why it works so well on the Internet, where memes are the currency of fame and power.