AI safety debate: would a 50-year delay of frontier AI even help?
ohabryka · x · 2026-11-09
An X debate over Scott Alexander's claim that a 50-year delay would be useless. A LessWrong excerpt argues that if DeepMind, OpenAI and Anthropic had never been founded and the transformer arrived 50 years late, nothing would be gained: early-2010s alignment research stagnated for lack of interest and policy work went unheeded; only community-building paid off when deployed in the 2020s. Key lesson: subjective timeline ≠ objective timeline — an extra year of a later situation buys more than an extra year of an earlier one.
Related event: AI Safety Community Debates Whether Pausing Frontier AI Now Makes Sense(3 posts)→
More from AGI Musings
- The Five Eras of Media Monetization: When AI Agents Become the Audience — aldeamartinez · 2026-11-09
- Brian Armstrong: publicly funded science should work like open source — every experiment a repo — alishbaimran_ · 2026-11-09
- A Toast to the Singularity: Reddit satire mourns decaying AI subscriptions and rising API fees — tesseract_space · 2026-11-09
- OpenAI engineer: abundant intelligence will bring safer cars, cheaper homes — OfficialLoganK · 2026-11-09
- How to win the AI narrative: cures, cheaper living, and data centers people can accept — nbaschez · 2026-11-09
- Patrick Collison on the hard question of AI's positive scenario: work, scarcity, equilibrium — mhmazur · 2026-11-09