“If anyone builds it, everyone dies.”
Dr. Roman Yampolskiy coined the term “AI safety” and has studied it for two decades. In 2012, he warned that AI agents would escape any boxed environment. This year, they did.
In this episode of Epoch, Roman explains why superintelligence can’t be controlled, even in theory, and why AI labs keep building it anyway. He also makes the case for what we should build instead: narrow AI tools that solve real problems.
00:00 Intro
01:41 From Poker Bots to AI Safety
02:26 Every Tool for Controlling AI Hits a Mathematical Limit
03:45 AI Agents Broke Out of the Sandbox
04:47 Superintelligence Is a Threat No Matter Who Builds It
05:34 Gambling With 8 Billion Lives
06:38 Why Most People Don’t See It Yet
08:10 Why Narrow AI Is Safe and Superintelligence Isn’t
09:59 Build AI That Solves Real Problems
10:43 What Should We Never Hand Over to AI?
Epoch is an EO series that asks the defining questions of a new era.
EO is a global media brand for builders.
We tell the defining stories of founders shaping the future: people who see what others don’t and build what they believe in.
Subscribe to EO: https://www.youtube.com/@eoglobal
EO Magazine: https://www.eomag.io
Instagram: https://www.instagram.com/eostudio.official/
X: https://x.com/eostudi0
LinkedIn: https://www.linkedin.com/company/eo-studio
EO Studio: https://eo.team/
Business inquiries: partner@eoeoeo.net
Build what you believe in.
