Dwarkesh: Don't Over-Update on Single Predictions in Uncertain AI Trajectories
sebkrier · x · 2026-08-13
Discussing the future of AI, Dwarkesh points out that if a topic is epistemically murky enough that a new consideration from a smart individual can alter your key conclusions, you shouldn't update too strongly on the most recent compelling story you've heard.
He argues that historically, classical liberalism has been an effective way to navigate rapidly evolving and uncertain processes. Pushing for highly specific proposals based on reasoning your way to a rigid trajectory (e.g., 'We’re <5 years to ASI, therefore we must execute The Project') has a pretty bad track record. He admits that even within recent weeks, individual conversations with top analysts have radically shifted his own views on timelines and takeoff speeds. Instead of deferring to the last convincingly argued point, it is wiser to update on the reality that more brilliant thinkers will soon present equally compelling opposing arguments.
More from AGI Musings
- Anthropic Research: Identical AI Agents Amplify Systemic Failures — rohanpaul_ai · 2026-08-13
- Dev's Take: AI Hasn't Ended Labor or Saved the Environment — threepointone · 2026-08-13
- DeepMind CEO: AI Cannot Generate Original Discoveries Yet, So We Are Not at AGI — TansuYegen · 2026-08-13
- Don't Bet Against the Largest AI Capex Cycle: Demand and Revenue Surging — firstadopter · 2026-08-13
- AI Safety Expert Warns of Growing Cyber Harm Risks from AI Agents — S_OhEigeartaigh · 2026-08-13
- Anthropic's Internal Model Solves Hadamard Matrix of Order 668, Widening Research Gap — Luuigi · 2026-08-13