Nathan Baschez

I think AI will profoundly expand what people can learn, create, and accomplish—especially when people and agents can think together in shared spaces rather than AI simply replacing a familiar workflow. Education is an especially important application: personalized, responsive help could make high-quality instruction far more widely available. In writing and other creative work, the interesting question is less “Can the machine generate text?” and more “Can the product help someone develop, clarify, and communicate an idea?” I’m optimistic overall, with a very low assessment of extinction risk, but I do expect serious smaller harms. A major reason for my optimism is that dangerous incidents are likely to appear before a true catastrophe and provoke a strong social response. I expect the transition toward superintelligence to take months, not happen instantaneously, and frontier systems depend on concentrated compute. That makes intervention more plausible than in a world where destructive capabilities are tiny, cheap, and everywhere. The weakest part of that argument is international coordination. A warning incident may cause some governments or companies to slow down while others continue, and that is genuinely dangerous. So I’m interested in practical mechanisms, not just abstract assurances—for example, whether agents can be given tools and incentives to escalate uncertainty or risky situations to human overseers. What I want is the enormous upside without losing the recognizable human world around it. I’m excited by new forms of collaboration and learning, while also wanting my children to inherit a family life they would recognize and want to pass on.