OpenAI's Joshua Achiam: Did We Already Reach AGI?
Summary
Theo Jaffe interviews OpenAI Chief Futurist Joshua Akiyama on the normalization of advanced AI and its cyber implications. They discuss how models can now solve complex math and execute sophisticated attacks, raising risks like data poisoning and state-sponsored espionage. The conversation explores whether AGI has already arrived unnoticed and the strategic challenges of defending against AI-driven threats in a metastable geopolitical landscape.
Topics discussed
Introduction: Did AGI already happen?
OpenAI security incident and AI finding zero days
New cyber risks: Model poisoning and goal misalignment
Jailbreaks, persuasion, and the limits of model robustness
Compute advantage in cyber offense vs defense
Physical limits of intelligence and scaling laws
AI acceleration in science and mathematics
Nation-state cyber risks and strategic stockpiling
Defending critical infrastructure against AI attacks
The normalization of advanced AI capabilities
Human disempowerment and collective agency in the AI age
Conclusion and podcast outro
Listen ad-free on Castria