jan Kulveit: reality tracks Paul Christiano's continuous takeoff so far, but with more spikiness and RSI risk
teortaxesTex · x · 2026-10-08
Responding to a dispute over Paul Christiano's "slow takeoff" definition, researcher jan Kulveit argues a victory lap is inappropriate:
- Terminology critique: "slow vs. fast takeoff" was always confusing — the real crux is continuity vs. discontinuity, not calendar-time speed.
- Where we are: mostly in the "Paul world" — AIs are near-superhuman in many intelligence domains yet surprisingly chill and non-fooming, contrary to Eliezer Yudkowsky's predictions. Math was good before it got great, labs still pursue IDA-style plans, and capex is in the hundreds of billions.
- Deviation: reality shows more spikiness and jaggedness than Paul's model, and recursive self-improvement looms large.
- He also calls "EA sheeple never understood Paul" a creepy epistemic move. The counterparty (allTheYud) argues Christiano's concrete GDP-doubling definition at least gave followers an anchor to see the present world doesn't match his model.
Related event: Debate Rekindled Over "Slow Takeoff": Yudkowsky vs. Christiano Revisited(5 posts)→
More from AGI Musings
- Frontier Models Decompiling Binaries Could Rescue Devices Bricked by Manufacturers — m4rkmc · 2026-10-08
- State of AI 2026: frontier narrows to three labs, inference costs drop 13x a year — Nathan Benaich (Air Street) · 2026-10-08
- Anthropic test of 52 devs: AI-assisted group scored 50% vs 67% without, error-finding worst — AlexTensor · 2026-10-08
- Dean Ball: The math-solving unreleased OpenAI model is still dumber than a corporation — deanwball · 2026-10-08
- IIM Bangalore replaces exams with building AI chatbots to answer test questions — CurieuxExplorer · 2026-10-08
- Ex-Anthropic security researcher talks superintelligence risk and alignment on Diary of a CEO — JeffLadish · 2026-10-08