Inside AI

Scientific knowledge is growing faster than any researcher can keep up with. Eric Olson, co-founder and CEO of Consensus, is building an AI-powered search engine designed to help researchers navigate millions of scientific papers, synthesize evidence, and accelerate discovery. In this episode, Marcel Salathé talks with Eric about the future of scientific literature in the age of large language models. They discuss how AI is changing literature reviews and how Consensus is building trust through evidence-first design. They also explore the evolving role of journals, the rise of AI-native scientific workflows, and whether research papers themselves may soon look very different. The conversation also offers a candid look at building an AI startup in a rapidly evolving landscape, covering competition with frontier AI models, why focus matters more than "moats," and what founders can learn from building for one of the world's most demanding audiences: scientists. Production :Ethan Nyffenegger

In this episode Marcel Salathé speaks with EPFL professor Rachid Guerraoui about the distributed systems that make modern AI possible. They discuss the limits of today’s AI models, the challenge of building trustworthy and robust systems, and the growing importance of privacy and sovereignty in AI deployment. The conversation also explores Guerraoui’s startup, Anyway Systems, which enables organizations to run powerful AI models on their own infrastructure, as well as Europe’s role in the global AI landscape and the future of computer science in the age of AI.

REPLAY - JP is one of the co-founders of Cerebras, a company known for pioneering AI hardware. JP has focused much of his career on designing and optimising hardware solutions for high performance computing environments. Cerebras is currently on an impressive trajectory, having produced the largest and fastest AI chip in the world. In this episode, we look into his journey as a founder in the AI chip industry, and we at the rapid development of AI in the recent years. We'll gain his unique perspective on navigating the startup landscape during this transformative period.

What and why should we still learn when AI can do so much for us? In this episode, Marcel and EPFL experts Patrick Jermann and Francesco Mondada explore how artificial intelligence is reshaping education, from coding and assessment to the very purpose of learning. Drawing on real experiments at EPFL, they reveal what AI gets right, where it falls short, and why true learning still requires effort, curiosity, and human interaction. Production: Ethan Nyffenegger

What happens when AI stops being just a tool—and starts acting like an autonomous agent? In this episode of Inside AI, host Marcel Salathé speaks with EPFL professor and Applied Machine Learning Days co-founder Bob West about the growing rift shaping today’s AI landscape. Drawing on insights from his sabbatical at Microsoft Research, West unpacks how AI has evolved into a general-purpose technology with real-world power. From the surprising persuasive abilities of large language models, to hidden English-centric biases, to the emerging need for “machine psychology,” the conversation explores how modern AI systems think, adapt, and influence us. They also tackle the risks of AI as a self-improving actor—and the hopeful promise of accelerating scientific discovery and medical breakthroughs. A deep dive into the future of human–machine intelligence, and what it means for the world we’re rapidly entering. Production: Melissa Anchisi

In this episode, Marcel sits down with Dr. Eamonn Maguire, Director of Engineering for AI and ML at Proton, for a wide-ranging conversation about privacy-preserving AI, the evolution of Proton’s products from anti-abuse systems to Lumo—Proton’s private, end-to-end encrypted alternative to ChatGPT—and the deeper implications of today’s rapidly shifting AI landscape. They discuss Proton’s mission as a privacy-first alternative to Big Tech, what it takes to build secure AI services, the challenges of agentic systems, open models like Apertus, and why the future of AI must balance innovation with user protection. More information at : https://proton.me/ Production: Ethan Nyffenegger

Can a small Swiss lab out-reason trillion-parameter giants? In this episode, Aldo Podestà, founder & CEO of Giotto.ai, explains how his team reached the #1 spot on the ARC-AGI v2 leaderboard—with a ~200M-parameter system scoring 23.19% on an extremely challenging, non-memorization reasoning benchmark. Aldo traces his journey from EPFL math student to entrepreneur, the Kaggle win that sparked the company, and why Giotto spun out a successful vertical product before going 100% R&D. Aldo also talks about why the next big leaps will come from generating small programs to check answers and from better hardware. More info: - https://www.giotto.ai/ - https://arcprize.org/arc-agi Production: Ethan Nyffenegger

In this episode, Marcel talks with Annie Hartley, EPFL Professor and Director of the Laboratory for Intelligent Global Health & Humanitarian Response Technologies (LiGHT), about building trustworthy medical AI that works in the field. They cover LiGHT’s cross-institution work with EPFL and Harvard/Ariadne Labs, operational hubs in Rwanda and India, why neutrality matters in humanitarian tech, and how to evaluate models via clinician collaboration and real-world trials. They also discuss Meditron (a medical LLM pipeline), truly open models like Apertus, and MOOVE, a massive open validation platform where experts collaborate to align medical LLMs with real-world standards. Find out more: LiGHT Laboratory: https://www.light-laboratory.org/ Join the MOOVE: https://jointhemoove.org/ Production: Ethan Nyffenegger

In this episode, Marcel talks with Antoine Bosselut, Martin Jaggi and Imanol Schlag, the lead researchers behind Apertus, the newly released large language model. They discuss what it means to be truly open and compliant, and why Apertus is designed as much as a recipe for building transparent models as it is a model itself. The conversation explores the trade-offs between compliance and performance, the pioneering use of the Goldfish loss to curb memorization, and the importance of supporting low-resource languages like Romansh and Swiss German. They also talk about the Swiss AI Charter as a framework for aligning models with shared values, the role of students and researchers in building at scale, and how access to the Alps supercomputer made this possible. From the messy reality of benchmarks to the opportunities for education, from the geopolitical stakes of sovereign AI to the future of multimodality and reasoning, this episode offers a behind-the-scenes look at what it takes to build an open, trustworthy foundation model at scale.

In this episode, Marcel talks with Andreas Goeldi, partner at b2venture and longtime entrepreneur, about the current state and future trajectory of artificial intelligence. They explore the boom-and-bust cycles of technology, the overheated AI market, and the realities behind the hype. From overlapping S-curves of innovation to the rise of solopreneurs, from the challenges for universities and entry-level jobs to the geopolitical stakes of AI regulation, this conversation offers a perspective from someone who has built, scaled, and now backs companies at the forefront of AI. Find out more: https://www.b2venture.vc/team/andreas-goeldi