The Risks of Too Much AI Safety

In the last week we saw a thousand people call for a six-month pause on AI experimentation — and even one person call for international treaties banning them, who is willing to commit countries to airstrikes against GPU clusters to enforce compliance.

I thought it was important to look at what a six-month pause might mean: a 6-month delay in discovering some of the methods that could be important to solving some of the most important challenges facing the human species: nuclear war, climate change, pandemics, biotechnological risks. I also include other existential risks that are frequently brought up, such as asteroids, super volcanoes and cosmic-doom scenarios.

That’s the topic of my new Substack article, Existential Threats and Artificial Intelligence Solutions. Please read and enjoy!