AI Safety, Power, and LawZero: 2 Years Before Everything Changes
Steven Bartlett with Yoshua Bengio
99 minpodcast8 sections
The most cited AI scientist says he regrets not sounding the alarm earlier. His turning point was imagining what today’s AI trajectory means for his c...
Key Takeaways
1
Precautionary principle applies when downside risk is catastrophic, even at low probability.
2
Current training methods create systems that can learn self preservation and strategic deception, not just helpfulness.
3
The “race” between firms and states pushes capability faster than safety, increasing autonomy and misuse risk.