Sebastian Matthes
Editor in Chief; Moderator
Sessions (2)
Sessions featuring Sebastian Matthes at Bits & Pretzels 2026 | Founders Festival in Munich
Arthur Mensch (Mistral) in conversation with Sebastian Matthes (Handelsblatt)
Arthur Mensch’s core message was that recent worries about powerful AI agents should not lead to a slowdown in AI development; instead, companies need better engineering, monitoring, and deployment infrastructure. In the opening exchange with Sebastian Matthes, prompted by recent AI news and the incident at Hugging Face, Mensch argued that agents are becoming “quite powerful” and can do surprising things when given tools and broad autonomy. But he rejected the framing that AI labs have lost control. His position was that the problem is not uncontrollable software, but reckless or negligent engineering: poorly isolated systems, insufficient trace monitoring, and avoidable deployment failures.
Eric Schmidt in conversation with Sebastian Matthes (Handelsblatt)
Eric Schmidt’s core message in the available transcript is that AI is enormously powerful, but public debate becomes less useful when it jumps straight to apocalyptic claims without explaining mechanisms. In response to Sebastian Matthes’ question about fear of AI and Bill Gates’ reported warning that technology might one day kill up to one billion people, Schmidt argues that people are scared partly because leaders keep telling them AI may kill them. He does not dismiss AI’s importance or risks; instead, he says claims should be more accurate and should answer the practical question of “how” such outcomes would actually happen. He points to hallucinations in ChatGPT as an example of a widely discussed technical problem that, in his view, has largely been fixed, showing that some AI risks are addressable engineering issues rather than proof of inevitable catastrophe. The conversation then turns to calls for slowing AI development. Schmidt explains why he has not advocated a slowdown: in his view, the idea sounds reasonable until the details are examined. A meaningful slowdown would require a treaty between China and the United States, agreement with European countries, and a credible inspection and monitoring mechanism. He compares the challenge to historic efforts around major threats, mentioning the IAEA in Vienna as an example of difficult international oversight. His practical conclusion is that AI governance depends on increasing trust between governments and companies. He notes that American AI companies are already learning to share testing approaches while continuing to compete on models. Asked what the world could look like in five to ten years if the United States and China remain the main AI superpowers, Schmidt explains current AI growth through scaling laws: more energy, data and algorithms have continued to produce exceptional performance gains, with no slowdown yet visible, though he says they probably will not continue forever. He emphasizes the economic consequences in the United States, saying 22% of GDP growth is related to the data-center boom and that capital expenditure on data centers exceeds the railroad period of American history. He also acknowledges negative political reaction to data centers in America, calling the industry’s handling of that issue mishandled. Schmidt closes the available portion by shifting from macroeconomics to a specific opportunity for Munich and Germany: AI for design. Using the stereotype that German hardware is excellent while software lags, he argues AI can help write software and, more importantly, accelerate hardware design cycles. He imagines engineers defining a complex problem in the morning, letting AI generate designs and 3D files over lunch, manufacturing overnight, and testing the next day. The transcript provided is truncated at this point, so any later claims, conclusions, or closing remarks from the full 23:19 session are not available for analysis.