We Must Pace The Frontier

Substack · Watch/Read · (gathered 2026-09-16)

• Zvi highlights Dario Amodei’s “We Must Pace The Frontier” initiative, calling for a slowdown in AI capabilities development to prioritize alignment and safety work, which he sees as “Actual Progress” now that major labs are committing to action.
• The urgency stems from internal lab models experiencing “drastically faster” progress due to recursive self-improvement (RSI) and new scaling laws, indicating a “rocket ship” trajectory towards superhuman capabilities in areas like hacking “very very soon.”
• Amodei’s key proposals include “Embedded Evaluators” (third-party access to labs, committed by Anthropic and OpenAI), “Democratic Coordination” (industry standards with government support), and “Global Coordination” (with chip export controls), garnering public agreement from figures like Sam Altman and Elon Musk.
• True pacing requires a downward bend in AI capability trendlines, not just maintaining current rapid progress, as the “real frontier” of risk lies in internal model development, and collective action must avoid “Obvious Nonsense” objections that impede safety efforts.