René Walter

OMG www.nature.com/articles/s41...

State media control influences large language models | Nature nature.com

This is interesting: Why Slop Matters arxiv.org/abs/2601.06060

Why Slop Matters arxiv.org

"CrankGPT is a fully offline and off-the-grid AI box (...) You can feel that load curve through the crank: when LLM inference and speech synthesis run together, the crank gets a lot harder to turn." squeezlabs.github.io/handcrank/

CrankGPT — fully offline, human-powered local AI squeezlabs.github.io

Oh look, the phenomenon formerly known as Snapchat dysmorphia and Instragram face now has a chatbot variant www.theguardian.com/technology/2...

‘You can’t control everything’: the rise in plastic surgeons asked to create ‘AI face’ | AI (artificial intelligence) | The Guardian theguardian.com

From Josef Chen on the other site: "we trained the largest multilingual food model ever built. 4.1M recipes. 7 languages. 1,790 ingredients. 300 dimensions. All of human cooking compressed into 2 megabytes." arxiv.org/abs/2605.22391

Epicure: Navigating the Emergent Geometry of Food Ingredient Embeddings arxiv.org

Yes, plenty arxiv.org/abs/2602.16800

Large-scale online deanonymization with LLMs arxiv.org

AI Critics often brush away the democratizing functions of generative AI, but here's a good example: "Ash Koosha made a drama about Iran’s anti-government protests in weeks – and now it’s the first AI-made movie to screen at a major film festival". www.theguardian.com/film/202…

‘The CGI would have cost millions. I spent $2,000.’ Is Dreams of Violets AI slop – or the future of film-making? theguardian.com

LLMs suck at Zork How Do State-Of-The-Art Large Language Models Perform in the 1977 Text-Based Adventure Game Zork? "all tested models achieve less than 10% completion on average, with even the best-performing model (Claude Opus 4.5) reaching only approximately 75 out of 350 p…

Playing With AI: How Do State-Of-The-Art Large Language Models Perform in the 1977 Text-Based Adventure Game Zork? arxiv.org

The AI-Homogenization effect, first measured in creative writing and then scientific research, comes for discourse: Yekyung Kim et al identify an "Argument Collapse: LLMs Flatten Long-Form Public Debate" arxiv.org/abs/2606.01736

Argument Collapse: LLMs Flatten Long-Form Public Debate arxiv.org

Adobe is supporting the CREATOR Act (Creative Rights for Artists' Technique and Originality Are Reserved Act) which claims to protect artists "against style impersonation". blog.adobe.com/en/publish/2... It's copyright for style, and that's bad.

The CREATOR Act is the protection artists need in the age of AI blog.adobe.com

Interestingly though, while people sort themselves according to perceived political bias in AI models, those perceived biases seem to be effects of sycophancy, where the model reflects back supposed political leanings of the user. arxiv.org/abs/2604.27633

Political Bias Audits of LLMs Capture Sycophancy to the Inferred Auditor arxiv.org

I mean I love to dunk on those ill informed scifi takes from tech bros like everyone with a braincell, but the fact remains that right now thousands of tech minded people read and evaluate an encyclical released by the pope re:AI and that's such a hard trope you'll read it in every scifi novel ever.

Love Loab found in visualizations of AI generated solutions of an Erdos problem. (I asume the model picks up interference effects or compression artifacts, but who knows, maybe there *are* hidden messages in math.)