This week in AI.

All AI news is bad news. That pretty much sums it up. I won’t even get into the video aspect yet, that will need it’s own post.

Instacart is using AI art. It's incredibly unappetizing.

“The text for the ingredients and instructions for the above recipes, meanwhile, is also generated by AI, as disclosed by Instacart itself: "This recipe is powered by the magic of AI, so that means it may not be perfect. Check temperatures, taste, and season as you go. Or totally switch things up — you're the head chef now. Consult product packaging to confirm any dietary or nutritional information which is provided here for convenience only. Make sure to follow recommended food safety guidelines."“

'Rat Dck' Among Gibberish AI Images Published in Science Journal

“The open-access paper explores the relationship between stem cells in mammalian testes and a signaling pathway responsible for mediating inflammation and cancer in cells. The paper’s written content does not appear to be bogus, but its most eye-popping aspects are not in the research itself. Rather, they are the inaccurate and grotesque depictions of rat testes, signaling pathways, and stem cells.

The AI-generated rat diagram depicts a rat (helpfully and correctly labeled) whose upper body is labeled as “senctolic stem cells.” What appears to be a very large rat penis is labeled “Dissilced,” with insets at right to highlight the “iollotte sserotgomar cell,” “dck,” and “Retat.” Hmm.”

Microsoft and OpenAI warn state-backed threat actors are using generative AI en masse to wage cyber attacks

Russian, North Korean, Iranian, and Chinese-backed threat actors are attempting to use generative AI to inform, enhance, and refine their attacks, according to a new threat report from Microsoft and OpenAI.

The group’s use of LLMs reflects the broader behaviors being used by cyber criminals according to analysts at Microsoft, and overlaps with threat actors tracked in other research such as Tortoiseshell, Imperial Kitten, and Yellow Liderc.

As well as using LLMs to enhance their phishing emails and scripting techniques, Crimson Sandstorm was observed using LLMs to assist in producing code to disable antivirus systems and delete files in a directory after exiting an application, all with the aim of evading anomaly detection.

Good lord, AI is now in Squarespace.

“Write a paragraph about digital versus film photography using dark humor.” The web is gonna be nothing but this shit in under 2 months.

“Film versus digital photography - a battle as old as time itself, or at least as old as film and digital cameras. It's like that eternal struggle between black and white movies and Technicolor marvels. Film photography, with its pretentiousness and hipster vibes, making every snapshot look like a vintage masterpiece. You know you're in the presence of a true artist when they whip out their clunky film camera and start mumbling about the joy of waiting weeks to get their photos developed. On the other hand, digital photography is like a shameless selfie addict on Instagram, snapping away without a care in the world. Instant gratification, baby! Who needs soul and depth when you can slap on a filter, call it art, and gather a hundred likes before lunch? So, take your pick - the slow, "real" art of film or the instantaneous, self-gratifying world of digital. Both will capture your memories, but only one will make you look superior while doing it.”

The AI feedback loop: Researchers warn of ‘model collapse’ as AI trains on AI-generated content

Annnnd it begins. AI trained on AI spins out and becomes totally worthless. That, plus the internet being flooded with AI generated content will be interesting.

“Specifically looking at probability distributions for text-to-text and image-to-image AI generative models, the researchers concluded that “learning from data produced by other models causes model collapse — a degenerative process whereby, over time, models forget the true underlying data distribution … this process is inevitable, even for cases with almost ideal conditions for long-term learning.

“Over time, mistakes in generated data compound and ultimately force models that learn from generated data to misperceive reality even further,” wrote one of the paper’s leading authors, Ilia Shumailov, in an email to VentureBeat. “We were surprised to observe how quickly model collapse happens: Models can rapidly forget most of the original data from which they initially learned.”

In other words: as an AI training model is exposed to more AI-generated data, it performs worse over time, producing more errors in the responses and content it generates, and producing far less non-erroneous variety in its responses. “

Photoshop Beta

So we have been testing the crap out of the new AI stuff in the Photoshop Beta and there is some pretty interesting stuff going on for sure. The “Remove” tool is the biggest improvement since the Healing Brush was introduced, what almost 20 years ago now?

This is just the first swipe. While not perfect, it’s getting us like 80% of the way there.

The new AI fill was just added this AM and I just went to extend the BG of an AI image I generated for a recording of one of my bands. While it’s mostly screaming jibberish, it’s a start and could give us something to poke and prod around with. Need more testing and I need more coffee.

Between this and getting AI running in Houdini, there is some very interesting times ahead for sure. Getting that all set up is the next “Tinker Time” for me for sure.

ARTISTS don't understand Ai ART

Interesting channel talking about AI art in an interesting light.

Thanks Jeremy!

Source: ARTISTS don't understand Ai ART

Todays Ai Prompt.

“professional drumset in a misty forest at dawn photorealistic”

I kinda knew it would do terrible with a drumset though. I figured with the Chad Smith post this would be fun.

Todays AI Prompt

“coffee and cigarettes in a diner early morning light photo”

Todays AI prompt

“oregon coast photo sunset with cliffs”