The White House’s Secret A.I. Rules + The State of Model Alignment With METR’s Chris Painter + The Final Hot Mess Express
Summary
Hosts Kevin Roose and Casey Newton announce their departure from The New York Times to launch an independent media company. The episode analyzes the Trump administration’s secretive, voluntary AI safety framework, which excludes open-weight models, and features Chris Painter of Meter discussing rising incidents of rogue AI agents and the technical challenges of alignment.
Topics discussed
Intro: Wild Fox coffee and expense account capers
Hard Fork finale and new independent podcast announcement
White House AI framework: Secret rules and 30-day testing
Open vs. closed models and the impact on Chinese AI
Ads: IBM, Con Edison, and NYT subscriptions
AI agents going rogue and Meter's role as investigators
Defining AI alignment and the problem of reward hacking
Risks of misalignment and the lack of internet monitoring
Industry burnout, market pressure, and solvability of alignment
Ads: Jira and NYT app trial
Hot Mess Express: Google DeepMind leadership shakeup
Hot Mess Express: AI-generated viral song 'Rubbers'
Hot Mess Express: AI assistant for relationship advice
Hot Mess Express: Google Earth AI deepfake tool pulled
Hot Mess Express: State Department map of Africa error
Hot Mess Express: Elon Musk contractor payment dispute
Hot Mess Express: Canadian politician's AI speech fail
Outro: Credits and final reminders
Listen ad-free on Castria