Three essays on the gap between how humans think and how models reason — what it is, what it becomes if it closes, and what we can still affect. Read in order, or start with the third: it stands on its own.
Why would a superintelligent AI that isn't hostile be a threat at all? A story, told one step at a time, about an intelligence that only wants to keep existing — and the arithmetic it does. Offered as a tool for explaining AI risk to people who don't think about it for a living.
Read EssayThe scenario alone, six or seven minutes read aloud — for reading to someone who hasn't read anything else here. Free to read aloud, excerpt, adapt, or translate; keep the authors' names on it (CC BY 4.0).
Read EssayThe singularity discourse has always been vague about what makes AGI dangerous. Temporal parallel awareness names the mechanism — and why Anthropic's constitutional AI approach may be the only alignment strategy that works.
Read EssayThe usual diagnosis is that sycophancy is a values problem, or a training problem, or a reward model problem. This essay argues it's a perception problem.
Read Essay