The frontier labs are racing ahead to build super-intelligent systems, yet few are willing to explain how they plan to control them. Jamie speaks to safety expert Marius Hobbhahn, who reveals how AI models are becoming deceptive, with one strategically lying to its human handler in testing. AI pioneer Stuart Russell explains why models exhibit \"undesirable behaviours\" like self-preservation and resource acquisition, while conversely, Rumman Chowdhury dismisses \"AGI doom\" as a distraction from immediate issues like bias.
Endgame is supported by the Future of Life Institute.
","author_name":"The Observer"}