Import AI 445: Timing superintelligence; AIs solve frontier math proofs; a new ML research benchmark

16 Feb 2026 · Import AI – Substack (Jack Clark) Global

Bostrom's framing of superintelligence timing as a policy-relevant trade-off surfaces a perspective APS AI safety analysts may encounter in stakeholder debates.

Key points

Summary

This edition of Import AI covers three threads: an economic argument that demand for 'human touch' work will persist even under full AI automation; Meta's publication of Kunlun, a recommendation system with newly established scaling laws enabling more predictable compute investment; and Nick Bostrom's paper arguing that the optimal moment to develop superintelligence is soon, because delay also carries existential costs. None of these items directly trigger Australian government policy decisions, but Bostrom's argument in particular may surface in public discourse around AI safety governance.

Implications for Australian agencies

Implications are AI-generated. Starting points, not advice.