Interesting paper. They took 10,272 writing prompts and gave each one to a human author and to five AI models: Claude, GPT, Gemini, DeepSeek, and Kimi. From these, they generated 61,608 stories, at around 5,000 words each to see if there was divergence at a narrative level. From the analysis, they found that they could identify a human-written story from an AI-generated one nearly 93% of the time. Left: Each story is a dot. The story’s 304 dimensional narrative feature vector has been projected down to two dimensions using Linear Discriminant Analysis (LDA). Brown dots are human. Right: Each story gets scored on how rare its narrative feature profile is, then mapped to a percentile: approximate 1.0 means highly unusual relative to the reference set. The models reliably produce narratively “average” stories, while humans more often into unusual territory arxiv.org/pdf/2604.03136

Oh this is good from Antón Barba-Kay - on how Magnifica Humanitas, alas, unwittingly plays into the technocratic paradigm which it condemns

While social media is polarising, evidence suggests AI may nudge people towards the centre. This holds true of all studied models. Grok is more right-leaning than other models, but also has depolarising effects. By @jburnmurdoch.