16:30-16:50 | Enabling Non-experts to easily and affordably create 3D Spaces
Bringing invisible and inaccessible spaces to life? That’s what digital design experts ART+COM and broadcaster Deutsche Welle (DW) are working on; an AI-powered tool that turns text, photos, videos, or audio into usable, realistic and mostly correct 3D models of real-world spaces. At DW, the tool primarily supports the investigative team, where reporters often rely on witness testimonies but lack visual evidence of what occurred behind closed doors. The tool enables them to make complex stories more tangible for audiences. With an iterative workflow and consistent scene generation, it produces output for web, VR, or video. Beyond journalism, it opens up applications in areas like spatial prototyping and design, where spatial concepts can be explored before being physically built.
16:50-17:10 | How Well Does Generative AI Support Journalists and Media Professionals in Text Content Production?
Large Language Models are increasingly used for content generation tasks such as summaries, social media posts, and news adaptations. Yet an important question remains: How well do they support professional journalists in producing high-quality editorial content?
This talk presents insights from a real-world study on AI-assisted text production conducted with professional journalists integrated into their daily workflow. It examines how generative AI affects editorial efficiency, output quality, and the overall production process. Also, lessons learned from comparing different model families in a newsroom setting are shared.
17:10-17:30 | Your Compliance Problem is bigger than you think – and AI can help Media Organisations navigate it
Regulations are multiplying. Press codes, editorial guidelines, broadcasting regulations, and new EU media legislation – the rules journalists and editors must follow keep growing. But reviewing content for compliance is slow, expensive, and requires expertise most editors don’t have in-house. Together with Deutsche Welle and dpa, Condat built an AI assistant that takes on this challenge. It reviews journalistic content against regulatory frameworks, flags potential violations by severity, and explains what went wrong — including the chain of reasoning, in plain language so that editors can understand the AI’s decisions. Learnings about making AI genuinely useful in high-stakes editorial workflows will be demonstrated.
17:30-17:50 | Confabulation Score: Identifying generated Bullshit
People who are asked for their opinion but have no idea themselves sometimes spout nonsense to make themselves seem interesting or to avoid looking foolish. AI-based text generation starts to make things up when it reaches the limits of its knowledge. Measuring such confabulations – statements that are both false and arbitrary – helps determine the reliability of generated texts and to explore system limits.
The project GenKI4Media is funded by the Federal Ministry of Research, Technology and Space as part of the initiative “KI-Innovationswettbewerb – Generative KI für den Mittelstand”.
This session will be held in German.
