Is AI Becoming Too Smart for Its Own Good? [audio]
AI expert Jonathan Cefalu discusses pressing challenges in AI safety, from prompt injection vulnerabilities to long-term risks of capable systems, including motivated reasoning, governability, and military applications.
NucleCast
Episode notes
AI expert, Jonathan Cefalu, joins the conversation to examine some of the most pressing—and underappreciated—challenges in artificial intelligence safety, from prompt injection vulnerabilities to long‑term risks posed by increasingly capable systems.
The discussion unpacks how seemingly technical issues like prompt injection can undermine reliability and security, why future AI systems may engage in motivated reasoning or resist human attempts to redirect them, and what “governability” really means as AI grows more sophisticated. Cefalu also explores the unique risks AI presents in military and defense applications, including the implications for deterrence, escalation management, and decision‑making under pressure.
The episode concludes with a forward‑looking discussion on re ...
... Read more