String theory is like crypto: it tackles hard problems that don't make sense from first principles, yet produces interesting techniques that might turn out useful for something else later.

Superintelligence is probably safe. A distributed network of AIs will be more intelligent than any single instance. Alignment won't be an issue either, as its emergent goal will be the same as ours: minimization of informational curvature.

Physicists are stuck in a local minimum, distracted by the elegance of Lie algebras and their symmetries, not realizing it's just an approximation.

Trying to explain how the universe works feels a lot like trying to explain Bitcoin to people.