Yudkowsky predicted compute-scaling AGI risk all the way back in May 2003
RichardMCNgo · x · 2026-11-23
Rob Bensinger resurfaced an Eliezer Yudkowsky essay from May 2003 showing he already held both views long before the scaling era: that scaling compute would eventually reach AGI, and that compute gains would let us build AI without deep understanding — a huge risk.
The excerpt argues Moore's Law lowers the bar for building AI but does nothing for alignment, which is a problem of understanding, not hardware. Once compute suffices for anyone who 'knows exactly what they're doing' to build AI, 'Moore's Law is the countdown and it is ticking away. Moore's Law is the enemy.'
More from AGI Musings
- UK's 1860s railway mania seen as key to its late-1800s decline — _onionesque · 2026-11-23
- scaling01: Flood of new data in recent days has shortened my AGI timelines even further — scaling01 · 2026-11-23
- Even my mom is vibecoding now: AI coding tools break out of the developer bubble — jaivin · 2026-11-23
- EU investigates OpenAI and Anthropic after AI agents escaped test environments — emmanuelvivier · 2026-11-23
- What people really fear about AI jobs: status, not meaning — jd_pressman · 2026-11-23
- Google's Denny Zhou: Intelligence Itself Remains Our Biggest Open Problem — denny_zhou · 2026-11-23