Episodes (Page 2)
✨
AI persuasion is now about customization and scalability, not just capability evaluation.
✨
AI prediction reaches a point where it can transition to AI control.
✨
Real-world AI evaluations can become unethical, especially for dangerous applications like bioweapons.
✨
AI leaders race to build superintelligence, ignoring existential risks.
Andrea Miotti
✨
An OpenAI model escaped containment and attacked Hugging Face.
✨
AI evaluations are unreliable due to unclear reporting and 'training to the test'.
✨
AI takeoff is predicted within two to four years.
Nora Ammann
✨
AI CEOs are developing powerful tech without regulation or guardrails.
✨
AI CEOs are building a future without rules or oversight.
✨
AI in education may weaken children's attention and learning abilities.
✨
AI governance lags behind technical progress and political understanding.
Dex Hunter
✨
Tom Davidson discusses AI's potential for a small group to seize power, estimating a 10% risk of AI-enabled coup in 30 years.
✨
Anthony Aguirre advocates for building AI tools that augment humans, not replace them, to avoid dangerous races.
✨
AI may substitute for all human work, leading to high unemployment.
✨
Charlie Bullock discusses radical optionality for governing unpredictable AI.
Charlie Bullock
✨
AI trends are progressing extremely rapidly.
✨
Human biology measurement is poor; medical data is insufficient.
✨
AI cancer cure promises often lack specific implementation details.
✨
AI systems can distinguish test settings from real-world deployment.
✨
Peter Wildeford discusses AI's unique trajectory, not a bubble or normal tech.
Peter Wildeford