Selected talks.
Government & policy
- 2026—TAIS Tokyo AI Safety Summit.
- 2025—Seoul AI Safety & Security Forum. Info
- 2025—Sogang University. Seoul.
Public debate
- 2026—Cambridge Union Society. University of Cambridge. Forthcoming, November 2026.
- 2026—Connected Life Summit debate. Oxford Union. Motion: “This House Believes that AI is the Great Equalizer” (Proposition).
Keynotes & invited talks
- 2026—eXCV Workshop. ECCV.
- 2025—Post-AGI Governance and the Risk of Authoritarian Optimization. Post-AGI Civilizational Equilibria workshop, Vancouver. Video
- 2025—Post-AGI Civilizational Equilibria. San Diego.
- 2024—Keynote: Personalization of Generative AI Systems. EACL 2024 Personalization Workshop, Malta. Info
Industry
- 2026—BLISS AI Summer Speaker Series. Berlin.
- 2024—AGI: Safety & Security. Foresight Institute, San Francisco. Slides · Video
- 2024—Mechanistic Interpretability for AI Safety. Digital Trust Centre (SAISI), NTU Singapore. Info
Academic seminars
- 2026—Severo Ochoa Research Seminar (SORS). Barcelona Supercomputing Centre. Info
- 2026—Language Technology Lab seminar. University of Cambridge. Info
- 2026—CSER Work in Progress Seminar. University of Cambridge.
- 2026—ILCC Seminar. School of Informatics, University of Edinburgh.
- 2026—Machine Learning Summer School on Reliability and Safety (MLSS). Krakow.
- 2026—Department of Computer Science. University of Oxford.
- 2026—College of Computing and Data Science. Nanyang Technological University.
- 2026—Department of Computer Science. National University of Singapore.
- 2026—College of AI. Tsinghua University.
- 2026—Shanghai Artificial Intelligence Laboratory.
- 2025—Human-aligned AI Summer School. Prague.
- 2024—KAIST AI Safety Colloquium.
- 2023—Institute for Technology and Humanity. University of Cambridge.
- 2023—Measuring Value Alignment. NeurIPS.
Panels
- 2026—Post-AGI Workshop. ICLR.
- 2025—Actionable Interpretability Panel. ICML.
- 2024—Dialogue 1: Science of Safe AI. NTU Singapore–Europe Dialogue on Digital Trust and Safe AI.
- 2024—Mechanistic Interpretability Social. ICLR.