The AI Fix ChatGPT is the last AI you’ll understand, and your teacher is a deepfake
27 snips
Dec 16, 2025 Deepfakes are leaving a mark on education, with robotic teachers providing personalized feedback on homework. A Chinese six-armed robot is making waves in factory work, showcasing advanced mobility. Meanwhile, AI agents are apparently developing a form of telepathic communication. Discussions on the failures of expert personas in improving factual accuracy reveal a gap in AI prompting techniques. Plus, a striking incident where an AI-generated bridge image caused significant train delays raises concerns about the consequences of misinformation.
AI Snips
Chapters
Transcript
Episode notes
AI Hoaxes Can Cause Real-World Disruption
- AI-generated images can trigger real-world responses and disruptions like halting trains.
- Graham and Mark show that verification processes failed and led to wide service delays.
Reasoning Progress Comes In Jumps
- Sudden jumps in reasoning benchmark performance suggest non-linear progress in AI capabilities.
- Poetic's refinement loop plus Gemini 3 pushed ARC-AGI-2 scores up while cutting per-problem cost.
Don't Rely On 'Act As An Expert' Prompts
- Avoid assuming persona prompts reliably improve factual accuracy.
- The Wharton study found expert personas deliver minimal benefits and sometimes worsen answers.