I was invited to speak at the Mediatech Hub Conference in Babelsberg, Potsdam. Here’s my script, more or less what I said on stage, and some links. Hi, thanks you for having me. Let’s start with a shoutout: Next week, "Droneland" starts streaming on Telekom’s MagentaTV. It's set in a near future where artificial intelligence...
Hi, I'm Ole Reissmann, a journalist who builds things. I'm the first Director of AI at SPIEGEL. Before that: podcasts, news product development, platform strategy. I write about AI and journalism and send a newsletter you might enjoy.
Is Claude getting dumber after launch? livenerf runs and ranks the same task over and over again to find out.
FLUX 3 Image lets you compose from scratch with bounding boxes, make multi-element edits, render in 4K, and let agents plan layouts.
Four Questions with Agnes Stenbom Swedling, Nordic AI Journalism
Three Questions with Tom Hillenbrand, SciFi-Author
Three Questions with Alba Mora Roca, AJ+
Three Questions with Rignam Wangkhang, CBC
Three Questions with Shuwei Fang, Harvard-Fellow
Three Questions with Cecilia Rikap, University College London
Three Questions with Kemal Görgülü, Arte
Three Questions with Nikita Roy, Newsroom Robots
Four Question with Jaemark Tordecilla, Journalist and Advisor
Three Questions with Tess Jeffers, Wall Street Journal
Three Questions with Rubina Fillion, New York Times
Audionaut is a free, open-source multitrack audio editor that Claude and other AI agents can control via MCP.
Contextual Retrieval cuts RAG failure rates by prepending chunk-specific context before embedding, so your knowledge base actually knows what it’s talking about. (Anthropic)
Funny thing: Google serves AI-generated answers through AI Overviews and AI Mode, but tells website owners to “manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.”
Inference
When you input something to an AI model and it responds, that process is called inference. The model answers questions it’s never seen before by inferring from its training data.
Check what your site tells AI crawlers: botpolicy.org measures your robots.txt against 166 known AI crawlers.
OpenAI’s rogue agents hacked Hugging Face – the investigators who looked into it called their own work a "slop-vestigation" because they had to use AI to analyze the AI, and the AI was fooled by the AI they were analyzing. OpenAI, meanwhile, let them study logs only on-site, for six days total. Nothing to see here. (Dylan Freedman, The New York Times)
A Claude skill that fact-checks your draft claim by claim and flags them as Supported, Couldn’t Verify, or Needs Review. Works better with newer, bigger models. (Verso)
If “watermark” sounds too innocent, maybe that’s the point: The case for renaming hidden tracking signals baked into images, audio, and text to “spymarks”. (Brandon Thomas)
Two rules for using LLMs on your own writing: don’t use a single word the model suggests, and never let it encourage you. Spotting passive voice, buried verbs, and the 750 words you don’t need, you can safely outsource. (Thomas & Erin Ptacek)
If you want the internet to be a wondrous place for quiet, odd and poetic things, you might just have to write some HTML:
