Haken dran – das Social-Media-Update der c't · Gavin Karlmeier

Hackt euch selbst! (mit Katrin Rönicke)

·1 hr 1 min·4 clips

A Gothenburg researcher invented a fake skin disease, uploaded two fabricated studies, and within weeks ChatGPT was telling users 90,000 people had it.

1. Haken dran, c't magazine's social media podcast, features host Gavin Kalmeier in conversation with journalist and podcaster Katrin Rönicke across a wide range of AI and platform topics. 2. Katrin Rönicke is introduced as a journalist, podcaster with her own shows including 'Wind und Wurzel,' and a guest specifically requested by Haken dran listeners. 3. The episode's central concern is the gradual erosion of the open, community-driven internet into algorithm-driven passive consumption — with concrete examples from medical misinformation research and AI hallucination experiments. 4. Rönicke opens by describing her relationship with social media as comparable to a post-divorce friendship: still maintained for professional reasons, but 'it never became what it was on Twitter,' where she first connected with others as a 'weird kid from the countryside' roughly 30 years ago. 5. A University of Duisburg-Essen study of 180 TikTok videos under hashtags including ADHD, depression, autism, anxiety disorder, narcissism, and PTSD found only 20% of videos had accurate information, with 20% outright false, 33% misleading, and 33% consisting of unverifiable personal experiences. 6. The hosts discuss a prior study in which researchers submitted emergency-level symptoms to ChatGPT Health and found it downplayed approximately half of cases that warranted immediate medical attention, suggesting patients 'lie down and relax.' 7. Rönicke offers a positive counterpoint: her two children, aged 16 and 19, are psychologically more self-aware than she was at their age, and she attributes part of this to social media communities normalizing mental health discussions. 8. Kalmeier argues the deeper structural problem is not the content itself but the passive consumption model: algorithm-driven feeds remove users' self-correcting reflex to seek out credible sources, whereas earlier hashtag-based discovery required active judgment. 9. Researcher Almira Osmanovic-Thunström at the University of Gothenburg invented a fake eye-area skin condition called 'Bixonimanie,' uploaded two fabricated studies to a preprint server in April 2024, and monitored how AI systems responded. 10. Within weeks, Perplexity reported that 90,000 people were affected by Bixonimanie, and ChatGPT began explaining the disease's purported causes and treatments — despite the condition having no basis in medical literature. 11. As of March 11, 2026 — nearly two years after the experiment — ChatGPT was still citing Bixonimanie as a real proposed subtype of periorbital melanosis (dark eye rings) caused by digital screen blue-light exposure, before reversing this answer briefly when challenged. 12. Penguin Random House is suing OpenAI over the generation of content from the 'Drachen Kokosnuss' children's book series, which was not licensed to OpenAI, in a case the hosts say could set a significant precedent for IP protection against AI-generated content. 13. Rönicke, a journalist, says ChatGPT-style conversational AI has reduced traffic to news websites because users now ask AI for summaries of current events rather than visiting news sources directly, which she predicts will ultimately be fatal for non-publicly-funded journalism. 14. A long New Yorker profile of Sam Altman, discussed by both hosts, portrays Altman not as erratic but as someone who 'knows exactly what to tell others to get what he wants' — and whom Rönicke describes as motivated primarily by power rather than money. 15. Rönicke argues OpenAI's original non-profit and open-source framing was a deliberate recruitment strategy that attracted idealistic top engineers willing to take pay cuts on the promise of preventing catastrophic AI, before Altman pivoted the organization toward profit. 16. OpenAI filed a request in California and Delaware for state prosecutors to investigate whether Meta and Elon Musk have engaged in anticompetitive behavior against OpenAI, a move the hosts say mirrors similar legal tactics Musk has used against competitors. 17. Greece announced a social media ban for users under 15, effective 2027 — with the announcement made by the Greek prime minister via a TikTok video, which both hosts note as an irony. 18. Rönicke and Kalmeier both argue that age verification systems tied to operating system wallets, Google and Apple accounts, or third-party services like Persona (whose first investor was Peter Thiel) create surveillance infrastructure that poses democratic risks beyond the child safety intent. 19. Listeners interested in AI hallucination research, social media regulation, and the economics of journalism will find the episode substantive and directly relevant. 20. Listeners who prefer English-language content or who want structured reporting without discursive tangents will likely find the conversational German format difficult to follow.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Haken dran – das Social-Media-Update der c't