Anthropic Can Now Read Claude’s Mind
Summary
This episode covers the UN’s call to ban autonomous weapons and Illinois’ new AI safety laws, alongside Alibaba’s legal victory against the Pentagon blacklist and China’s crackdown on anthropomorphic AI agents. Market news includes NVIDIA’s rumored server delays and Nemotron’s download milestone. The main segment analyzes Anthropic’s research on “J-space,” a tool that reads Claude’s internal reasoning to improve model safety and performance.
Topics discussed
Intro, sponsors, and show overview
UN calls for ban on killer robots and AI governance
Analysis of UN dialogue and regulatory evolution
Illinois signs strict AI safety and audit bill
Alibaba sues Pentagon over defense blacklist
China bans AI anthropomorphic chatbot features
Mercor hits $2B revenue in AI data industry
NVIDIA server delays and semiconductor market shifts
NVIDIA Nemotron open model hits 100M downloads
Sponsors: KPMG, HyperAgent, Robots and Pencils, Blitzy
Anthropic research: Reading Claude's internal thoughts
Global Workspace Theory and the J-lens tool
Five properties of the model's private workspace
Safety implications and training internal reasoning
Neuroscientist commentary and consciousness debate
Listen ad-free on Castria