Superintelligence: Paths, Dangers, Strategies · chapter 4 · id superintelligence-c4-01-an-ai-system-could-rapidly-sel

“An AI system could rapidly self-improve from roughly human-level intelligence to vastly superhuman intelligence in days or weeks -- a 'fast takeoff' or 'FOOM' scenario.”

needs contextconfidence: high⚠ extracted by pipeline, re-audit pending

Receipts

Hanson-Yudkowsky AI-FOOM Debate (2008)source alive

Yudkowsky argues a single key algorithmic insight could produce a discontinuous jump; software can be copied instantly; the search space narrows once you have a smart-enough searcher.

Kaplan et al. 2020, Scaling Laws for Neural Language Models (OpenAI)source alive

Loss scales as a power-law with model size, dataset size, and compute. Capability improvements are smooth and predictable, not discontinuous.

AI Impacts: Discontinuous Progress Investigation (Katja Grace)source alive

Investigated 37 technological trends. Most don't demonstrate discontinuity. Big jumps equivalent to >100 years of progress are rare (10 events found). Default expectation should be continuous progress.

Alignment Forum: Yudkowsky and Christiano Discuss Takeoff Speedssource alive

Christiano argues AI capabilities will be deployed incrementally across the economy, each marginal improvement yields diminishing strategic advantage, transition takes years not days.

This claim is a stable, citable object. If you can falsify a verdict, tell us — corrections are loud here.