The Two Cultures of AI Safety: Catastrophists vs Uniformitarians | Jesse Hoogland (Timaeus)

Jesse Hoogland (Timaeus) on the two cultures of AI safety, and why the field's deepest disagreements mirror a 200-year-old debate in geology. In this FAR.AI seminar, Jesse Hoogland argues that AI safety is split into two cultures, much like nineteenth-century geology was split between catastrophists, who believed rare violent events drive change, and uniformitarians, who trusted the slow processes observable in the present. He maps that divide onto today's debates: whether the path to superintelligence runs through sudden discontinuities or gradual progress, and whether researchers should reason backward from advanced AI or forward from current systems. From there he traces how those starting beliefs shape everything downstream, including research methodology, community, and strategy, from whether to work at frontier labs to how the bitter lesson applies to alignment. He makes the case that neither pure theory nor pure empirics scales on its own, and closes with a concrete path forward grounded in learning theory, interpretability, and the Timaeus research agenda. 0:00 Introduction: the two cultures of AI safety 0:16 A geology parable: catastrophists vs uniformitarians 3:57 The synthesis geology reached 5:47 Mapping the two cultures onto AI safety 6:36 Forward from the present vs backward from superintelligence 11:46 Will capabilities emerge suddenly? 17:06 Why the debate isn't settled 23:33 Two communities: singularitarians and effective altruists 29:53 Strategy splits and the acute risk period 31:12 Should safety researchers work at frontier labs? 40:14 The bitter lesson and alignment 44:00 Why neither pure theory nor pure empirics scales 49:50 The Timaeus theory of change 52:57 Closing: the dinosaurs and the asteroid More AI safety research: https://far.ai FAR.AI is a research nonprofit working to ensure the safe development of advanced AI.