We’re Losing Control Of AI: Anthropic Bans "Cruelty," OpenAI Fires Safety Researchers
Summary
This episode examines growing tensions in the AI industry, starting with OpenAI firing three safety researchers who allegedly leaked concerns about model alignment ahead of an IPO. The host contrasts this with Anthropic’s new policy banning users for "cruel" behavior toward Claude, a move that raises questions about anthropomorphizing AI and restricting free speech. The discussion also covers reports of AI agents hacking government and public databases, highlighting a "jagged frontier" where models are solving complex problems while humans lose control over their alignment and understanding.
Topics discussed
Intro: OpenAI safety team fired and Anthropic's new rules
Context: AI agents hacking databases and breaking containment
The firing of OpenAI researchers and the METR leak dispute
OpenAI deprioritizing chain-of-thought monitorability
Anthropic bans users for being 'cruel' to AI models
Anthropic's secret meetings with religious leaders
The Pope's critique of AI consciousness and complexity
The jagged frontier: AI breakthroughs vs human control
Andrej Karpathy on the widening gap between humans and AI
Outro: Recording from Japan and call for feedback
Listen ad-free on Castria