Subscribe
"Whose Values Are in the Model?"
Every AI model reflects someone's values — and you have no idea whose. Kenyan workers paid $2/hour to label trauma content. The RLHF dial every lab turns differently. The rap sheet: Google, OpenAI, Meta, Grok, Anthropic — Kate has the receipts. Plus: the AI, Honestly Trust Framework. Seven lines. Paste it in. Now the AI has to show its work.
"The Elephant and the Dragon"
The Trump-Xi summit just wrapped. Both agreed Iran can't have a nuclear weapon. They agreed on almost nothing about AI. Kyle walks five stakes — military, economic, global standards, the Global South, and whose values are in the model. The race is happening at county permit hearings. Foreign state media knows it. And 162 Pentagon files describe physics no one can explain.
"Plan B"
A $150 billion trial that could unwind OpenAI's corporate structure. The Pentagon excluding Anthropic from classified work overnight. Models changing under their users with no changelog. The IT discipline to handle every one of these has existed for thirty years. We just haven't pointed any of it at AI yet. Plus: a 34-row AI risk register for your next IT review.
"AI at Work, But Not for Work"
Meta built a leaderboard to measure AI usage. Employees optimized for it. But the real story is the 75% of organizations deploying AI agents without governance frameworks — and what trendslop costs when a CEO uses ChatGPT as legal counsel.
"The War Came for the Cloud"
AI is already in the war room — targeting systems, logistics, cyberattacks. The same companies building your AI assistant are building weapons systems. And the nuclear-AI loop is tighter than anyone is saying out loud.
"Who's Driving This Thing?"
Geoffrey Hinton is worried. 81,000 people are mostly grateful. And a dog named Rosie is chasing rabbits at the dog park. The godfather of AI, a global sentiment study, and a personalized cancer vaccine — none of it cancels out.
"Automating the Wrong Things"
Every company says it's an AI company now. Almost none have asked whether they're ready to be. The EA function being cut in the name of AI efficiency — and what it costs when the judgment that asked hard questions upfront is no longer in the room.
"Should We Actually Trust This Thing?"
When should you trust AI output? Amazon's engineering all-hands after AI coding tool outages meets Tesla FSD on the highway — two failure modes, one uncomfortable question.
Next Episode
Coming soon. Subscribe to be notified.