Where's the "intelligence explosion"?

Summary

Argument essay on why no runaway intelligence explosion has materialized. Distinguishes five forms of AI self-improvement, argues current self-improvement feedback would need to be roughly 5-10x stronger to be self-sustaining, and expects narrow superintelligence in verifiable domains while seeing no current evidence for runaway Type-5 recursive self-improvement.

Keywords

recursive self-improvement · AI forecasting · essay

Topics

recursive self-improvement, AI forecasting, essay

Research notes

• Discovery: Shared in #random-papers as a Noahpinion (Noah Smith) Substack link; guest essay by Ramez Naam.
• Key findings: Points to gaps between coding-benchmark/forecast performance and internal autonomous research-task performance as evidence that headline capabilities overstate self-improvement.
• Limitations: Opinion essay, not peer-reviewed research; argument rests on the author's interpretation of internal-task evidence.
• Pairs with the recursive-self-improvement theme; read alongside the ROFT and SPSD papers in this batch.