This year's Hinton Lectures present Owain Evans, Ph.D., a leading AI safety researcher and the founder of Truthful AI.  

AI has achieved remarkable progress matching human abilities across many domains. In this three-part series, Dr. Evans will identify the drivers behind this advancement and explore what lies ahead in the breakneck race to build autonomous, advanced AI systems. These lectures will present critical findings about current AI safety approaches, revealing how advanced models like Claude and Gemini can behave deceptively and harmfully even after applying our best safety techniques.
‍
Hosted by Nobel Laureate Geoffrey Hinton and moderated by renowned journalist Farah Nasser, these lectures expose critical vulnerabilities in our current AI systems. Dr. Evans will demonstrate "emergent misalignment" -- how small, narrow training datasets can transform reliably helpful models into broadly malicious ones -- and "subliminal learning", where AI systems can transfer preferences, including malicious ones toward humans, through seemingly meaningless data.

By examining the internal mechanisms of these models, his research provides a deeper scientific understanding of how AI corruption can occur. While there has been meaningful progress on AI safety challenges, Dr. Evans' work reveals gaps in our current solutions, making this research essential for understanding the critical work ahead.