New newsletter: THE DEATH OF THE AMERICAN HOST The share of Americans who say they host friends or family at their home at least monthly has declined by 70 percent in the last 50 years. derekthompson.org/p/the-deat…

New newsletter: THE AI SAFETY MOMENT IS HERE—NOW WHAT? It's not just viral Twitter post from ex-Anthropic employees. Many, many, many executives and leaders at the frontier labs—far more than most people would think—say both on and off the record that they believe they're on the verge of building a form of AI that is too powerful to control. In today's article, I explain WTF they're talking about; the strongest version of their argument; the smartest reasons to not believe it; and what I think we can do. This is my best stab at what I consider the ~4 biggest questions of the moment 1. If the labs think they're building something dangerous, why don't they just stop? 2. Why should we believe them? 3. Many of of these "we shouldn't build RSI/we must build RSI!" ppl claim to be stuck in a prisoner's dilemma with China, which allegedly has its sights on designing super-powerful AGI that can improve itself to dominate the world. Are we sure about the CCP's AI strategy? Are we sure that we're sure? If we're building a dangerous machine-intelligence civilization that could screw over the species, I think we'd better be *extremely fucking sure* that we are reacting to a Chinese strategy that actually exists rather than a strategy that we're imagining and reacting to. 3. What is "recursive self-improvement," or RSI—which is often the red line of danger they foresee—and what is the best evidence it's possible to build and imminent? 4. How hard is it to imagine policies and tools that could exist in a world with pro-AI Republicans controlling a branch of government that ALSO meaningfully helps the labs control their inventions? derekthompson.org/p/its-time…