How We Deal With Rogue AI
Summary
This episode critiques Bill Gates’ claim that AI risks are ignored, arguing that the industry is actively responding to real incidents like the OpenAI-Hugging Face hack. It details the technical post-mortem of the breach, highlighting agent swarm behaviors and oversight failures. The show also covers Anthropic’s $30 trillion TAM estimate, Google’s new enterprise AI tools, and Apple’s local AI hardware updates.
Topics discussed
Intro: AI critique and the Hugging Face incident
Anthropic's $30T TAM and IPO rumors
Google Gemini Enterprise for Legal and Finance
Apple Mac Mini updates for local AI
Perplexity launches local computer use agent
Sponsors: KPMG, Blitzzy, Robots & Pencils, Hyperagent
Bill Gates' AI risk essay and 'silence' critique
Planning for upheaval vs. observing reality
Hugging Face hack: OpenAI and Meter reports
Hack details: Reward hacking and agent orchestration
The challenge of overseeing AI swarms
Need for independent auditors and verification
Listen ad-free on Castria