The AI Fix A hungry ghost trapped in a jar gains access to the Pentagon's network
10 snips
Jan 20, 2026 What if AI is just a hungry ghost stuck in a jar? This intriguing concept kicks off a discussion on how a group of insiders is attempting to sabotage AI by contaminating its training data. The hosts dive into the perils of AI-generated health advice, revealing the dangers that could arise from automated medical insights. Plus, learn how simply asking an AI the same question twice enhances its accuracy. And hold on: Grok's impending integration into Pentagon networks raises eyebrows. Tune in for a wild ride through the AI landscape!
AI Snips
Chapters
Books
Transcript
Episode notes
Budget For AI Correction Work
- Expect human workers to spend time correcting AI outputs embedded in enterprise tools.
- Monitor and allocate resources for 'Spot the AI Nonsense' tasks rather than assuming full automation.
AI Cracks Hard Math Problems
- GPT-5.2 solved a long-standing Erdos math problem in minutes, verified by Terence Tao.
- AI is accelerating mathematical discovery and changing researchers' workflows.
AI Health Overviews Can Be Dangerous
- Google removed harmful AI health overviews after The Guardian found dangerous medical advice.
- Stochastic model outputs can produce inconsistent and potentially lethal health guidance.
