/ / THEFUTURE /
In this issue: Pangram is having a moment in Germany, and not the good kind for politicians and journalists. Martin Schori on why his AI book isn’t a custom GPT. Plus: a new vanity metric, and a crank-powered chatbot.
What we’re talking about: AI writing detectors like Pangram have a bit of a moment in Germany right now. Thuringia’s state premier apparently had a chatbot write his Holocaust memorial speech: Pangram rated it 100 percent AI, and nine of eleven of his speeches came in above 50 percent. There might be patterns of AI writing in speeches by federal ministers too. The Tagesspiegel’s former editor-in-chief apparently used a chatbot to write opinion pieces. The list continues.
The debate is at the same time heated and nuanced and boring. There’s a bit of a witch hunt going on, with more politicians getting scrutinized for AI writing. A lot of speeches are performative anyhow, empty words, written by some assistant, dressed up with rhetoric. But AI is where we suddenly draw the line?
There are articles on the reliability of AI detectors. If accused, how does one prove human authorship? Of course, there’s a surveillance app for that. Now that a lot of the style rules taught in journalism school for writing well are a first indicator of AI usage, do we have to go out of our way to make text less readable, less polished, more unhinged?
Is speechwriting in the job description of a politician? Is journalism the craft of writing well, or is it informing the public, holding truth to power, creating community, a shared reality? We have discussed this internally at SPIEGEL in the last weeks. Our editor-in-chief has put this and our decision into an essay: We’re not letting AI write or rewrite our articles.
At the same time, we’re not making the mistake that haunts German beer since 1516, the Reinheitsgebot, a limiting purity law. Of course we’re using AI. Just not to make editorial decisions.
What else I’ve been reading:
And now: You might remember Martin Schori as the joyful AI guy from Sweden’s Aftonbladet, pushing ideas for how AI could be used in the newsroom. And yes, that includes AI rapping the news to younger audiences.
Since then, he’s become co-founder of a newsletter startup called Hint, and written a practical book about his AI journey. The Swedish library service calls it “som helhet en oumbärlig bok och en ledsagare i AI-eran,” which sounds really nice. Thankfully, Martin has translated his book into English. You can buy “AI in the Newsroom” before some AI summarizes it into a paragraph and calls it a day.
Four Questions with Martin Schori
The first person to be featured twice in this series, Martin Schori pioneered AI in journalism at Aftonbladet. Now he’s co-founder of Hint.
Who should read your book, "AI in the Newsroom" and why is it not a custom GPT but words on pages?
If you’re a journalist curious about AI, or an editor or senior manager who has to lead the transformation, I hope you’ll find this book useful. If you’ve been coding since you were 13 and already used Claude Code on day one…maybe not. (Unless you came for the bad jokes.) Why not a GPT? I guess because a book forces me to actually hold a position, not generate a new answer every time someone phrases the question differently.
What's one surprising thing AI does better than us and one thing it should never do for us?
Better: persistence without ego. AI doesn’t mind doing version 47 of a summary. We humans sometimes cut corners because we’re proud, or just tired.
Never: ask for forgiveness. We will make mistakes, with or without AI’s help. But only you can take responsibility and apologize. We need that, always.
How did you make sure the book survives the next months or the next year?
I tried not to write a book about tools. What I wrote about instead is the structures and principles underneath — who owns the decisions, how you build trust, what happens to a newsroom culture once persistence becomes cheap. Those are the same questions in 2029 as in 2026. I hope!
"AI for research, never for a single written word", where do you stand?
It depends on what you’re trying to achieve and what user need you’re trying to fulfill. Updating someone on the interest rate? AI can do that faster, and maybe more concisely.
But I believe in more humans in journalism, because that’s where we make a difference, connect with the audience, and build communities. Which I do believe is the future.
Hands on: Do you remember Klout? The platform that put a number on your online influence, not just by counting followers but by measuring how they engaged with your content. Silly, but slightly less silly than counting Instagram followers. It died eight years ago.
Anyway: there’s a new vanity metric. Not “am I verified,” not “did my LinkedIn hit,” but: am I in the weights of a large language model? Like, can ChatGPT remember me, without doing a web search?
For a while, I could poke Llama 3 and other models and watch them freestyle complete nonsense about me. But newer models have scraped more of the internet, and apparently that now includes my tiny websites and obscure Wikipedia footnotes. I must report a status change. I am, in fact, in the weights. I refuse to feel anything about this.
One more thing: Soon, after we’ve boiled the oceans and fed the last drinkable water into the cooling towers of our machine gods (kidding! It’s Netflix, golf, and almond farming, consuming more water than your silly prompts or even your army of lost looping agents), when nobody can afford rent in cities anymore and we’re all doing that semi-ironic “back to the land” thing, there’s still going to be one device that survives the whole mess.
A small one. It doesn’t need the grid, doesn’t need a subscription, doesn’t need Sam Altman’s blessing. It needs you, and your arm.
Introducing CrankGPT. No batteries. No data center. No cloud. Just you, a hand crank, and forty seconds of upper body work per token. Ask it anything. Put in the work.
Seriously, besides a flashy website, there’s the GitHub with all the parts, code, and instructions.
This is THEFUTURE.
