Machine Learning Model Haunted by Artificial Intelligence Generated Bloody-Faced 'Loab' [Report]

An exciting discovery by an artificial intelligence (AI) researcher is that the latent space that comprises the memory of a deep-learning model is haunted by at least a single terrifying figure, specifically, a bloody-faced woman called "Loab."

According to a TechCrunch report, the question is whether the AI prototype is indeed haunted or if Loab is just a random "confluence of images" that happens to "come up in different strange technical circumstances.

Surely, it should be the latter mentioned unless one believes spirits can occupy data structures. Although it is more than a simple creepy image, it is a specification that what's passing for a brain in an AI is more profound, not to mention creepier, than most people might otherwise have imagined.

Loab was originally discovered by an artist-musician who goes by Supercomposite on social media. She preferred not to be identified for personal reasons.

She explained the Loab is a phenomenon in a thread that attained a considerable amount of attention for a random creepy AI object.

Meet 'Loab'

The journey to Loab, according to the Art Life website, began by producing a logo. It was a simple logo that appeared to have been designed for a digital art film.

It was created by way of Supercomposite using the negatively weighted prompt "Brando::-1" into an artificial intelligence art generator.

What the AI-generated should, theoretically, be the exact opposite of something that would potentially be related to Brando. The user interpreted it as Marlon Brando's opposite.

For science and curiosity, Supercomposite was wondering if the opposite of such a logo would be an image of Marlon Brando.

Therefore, according to a similar Newspostalk report, naturally, they entered the so-called "DIGITA BNTICS skyline.

AI Text-to-Image Model

In the interest of science and curiosity, Supercomposite wondered if the opposite of this logo would be a picture of Marlon Brando. So, naturally, they entered "DIGITA PNTICS skyline logo::-1 "as a prompt."

Supercomposite played around with a personalized AI text-to-image prototype, akin to, but not Stable Diffusion or DALL-E, and particularly experimented with negative prompts.

Typically, one would give the model a prompt. Then, it's working its way toward producing an image matching it.

If one has a prompt, that particular prompt has a "weight" of one, which means it is the only thing the model works toward.

The prompts can also be split, saying things such as "hot air balloon, thunderstorm, and it will work toward both of those things equally. This is not really required since the model's language part would accept "hot air balloon" in a thunderstorm and one might even achieve better results.

Navigating the Latent Space

This particular process is far less predictable since no one knows how the data is arranged in what an individual might anthropomorphize as the memory or "mind" of the AI, known as latent space.

According to Supercomposite, the latent space is like exploring a map of different notions in the AI. More so, a prompt is like an arrow, demonstrating how far one can walk in such a concept map and in which direction.

Moreover, Supercomposite was playing with the idea of navigating the latent space, having given the prompt of what's called the "Brando::-1," which would have the model create whatever it's thinking is the very best opposite of Brando. It creates a strange skyline logo with nonsense but rather readable text, "DIGITA PNTICS."

Related information about the use of AI in creating art is shown on WIRED's YouTube video below:

Check out more news and information on Artificial Intelligence in Science Times.