DeepMind Institute essay maps AGI to ASI: effective compute grows at least 10x per year
daniel_mac8 · x · 2026-11-20
Shane Legg announced a new DeepMind Institute essay examining pathways from AGI to ASI (Artificial General Super Intelligence).
Its "cognitive industrialization" section compounds three drivers: hardware gains (1.5x compute per dollar yearly for six decades), algorithmic efficiency (training compute for fixed performance falling 3-6x per year; inference efficiency up 3x yearly), and investment (AI hardware spending 2.5x per year; largest training runs >4x yearly). Multiplied together, global effective compute grows at least 10x per year — 10 billion times over a decade, enough to run more equal-quality agents than there are people on Earth. Other chapters cover four AGI→ASI pathways and recursive self-improvement.
Related event: DeepMind Institute Outlines Paths from AGI to Superintelligence(6 posts)→
More from AGI Musings
- Google researcher ibab argues LLM pretraining distills humanity's unpaid collective work, and charity can't settle the debt — ibab · 2026-11-20
- Garry Tan backs Bostrom's 'if nobody builds it, everyone dies', urges taking Deep Utopia seriously — garrytan · 2026-11-20
- Superforecasters put 33% odds on a week-long outage hitting 1M+ people in an OECD country by 2028 — NathanpmYoung · 2026-11-20
- Sentinel brief 'Consider Cyberarmageddon' extrapolates cyberattack growth to societal failure — NathanpmYoung · 2026-11-20
- Google researcher ibab: models trained on humanity's commons should be open-sourced — ibab · 2026-11-20
- $1M into a public benchmark gets labs to spend $100M+ hillclimbing it — _jasonwei · 2026-11-20