A review of Nick Bostrom's 2014 book 'Superintelligence: Paths, Dangers, Strategies', evaluating its predictions against today's AI landscape. The reviewer finds many predictions remarkably accurate — including AI mastering Go, compute costs rising, and AI as a psychological manipulator — while noting some misses, such as natural language understanding not leading to general intelligence. The review praises the book's core argument that safety guardrails must be built before superintelligence arrives, draws parallels to nuclear non-proliferation, and critiques the book's over-reliance on figures like Eliezer Yudkowsky and the now-compromised Effective Altruism movement.
Sort: