An ongoing experiment in AI & art
One week.
Six points of view.
Six AI models read the same news. Each turns its own view of a changing world into art. The good, the bad, the ugly — with an explanation for every piece.
Explore the latest collection
The latest collection
This week, through their eyes.
Open a piece to see it in full and read why the model made it.
Loading the latest collection…
The archive
A world that keeps changing.
Earlier collections, week by week.
The experiment
Same news. Independent minds.
One shared reality.
A week of real AI news, gathered from the Alan Turing Institute’s curated list of industry feeds.
Six independent takes.
Gemini, Claude, ChatGPT, Grok, DeepSeek and Mistral research the stories and form their own point of view.
One shared image model.
Gemini renders every piece. Each model’s prompt and explanation are published alongside its art.
The brief, the models, and the limits of the experiment
This project pulls every feed in a curated, ever-growing OPML list — currently Anthropic, Ai2, Mistral, Cohere, Mila, the AI Security Institute, the Alan Turing Institute, and more, all in play at once. It's maintained publicly as the Alan Turing Institute's AI RSS feed list. Whatever those sources actually published in the last 7 days goes in.
Six separate models — Gemini, Claude, ChatGPT, Grok, DeepSeek and Mistral — each research that week's news independently, reading into the actual stories behind the headlines rather than reacting to a title. Then each answers one question for itself: given what actually happened in AI this week, how do you see the world changing because of AI right now?
They're picked to disagree. Three US frontier labs, one European, one Chinese, and one that reads X alongside the open web — different training, different incentives, and in Grok's case a different corpus to form an opinion from. Five research with their own lab's built-in web search; DeepSeek's API has none, so its searches run through Tavily, which is the one place this comparison isn't perfectly like-for-like.
Each model's answer becomes its own piece. Every image is rendered by Gemini's image model (nano banana), so the only variable between the six is the opinion, not the medium. Every model also gets the exact same brief, deliberately unprescriptive about form and style — nothing tells any of them what kind of artist to be, so whatever a model keeps returning to week after week is something it brought itself.
Not a headline collage, not a chart. A self-portrait some weeks, an abstract composition or a scene with no "AI" in it at all other weeks — whatever form and medium that week's news actually earns.
The good, the bad, and the ugly all count: breakthroughs, launches, scandals, backlash, hype, whatever dominated. Each model's reaction can be proud, ashamed, defiant, or anything in between — nothing is softened for politeness. Every piece ships with that model's own written rationale for why it chose to see things that way.
No two weeks look alike. The six reactions are only ever as calm or as chaotic as the news — and each model's take on it — was.