A genealogy of AI safety: how directions are born, and how they die
LessWrong mapped 20 years of AI safety research (2005–2026) across 323 events, 129 actors, and 18 research directions.
The standard narrative—philosophers raise concerns, then LLMs arrive and alignment becomes central—flattens a messier reality: directions emerge, some merge into the mainstream, others fade, and funding lags research momentum by years.
The genealogy reveals how attention and money rarely move in sync, and which bets compound versus which disappear.