Import AI 445: Timing superintelligence; AIs solve frontier math proofs; a new ML research benchmark
Tracks live debates on AGI timing and AI research autonomy - background context for APS teams monitoring frontier AI risk narratives.
Key points
- Import AI 445 covers superintelligence timing arguments, frontier math benchmarks, AI research agents, and Meta's recommender scaling laws.
- Nick Bostrom's paper on optimal AGI timing argues swift development with a potential late-stage pause is preferable to prolonged delay.
- Limited direct APS operational relevance; useful as a signal of current frontier AI research and safety discourse directions.
Implications for Australian agencies
- Monitor Teams tracking frontier AI safety discourse may want to monitor the Bostrom paper and associated debate, as arguments about AGI timing are likely to surface in international policy discussions.
- Monitor Agencies interested in AI capability evaluation could note the First Proof and AIRS-BENCH benchmarks as indicators of how the field is approaching autonomous AI research agent assessment.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
"Import AI 445: Timing superintelligence; AIs solve frontier math proofs; a new ML research benchmark"
Source: Import AI – Substack (Jack Clark)
Published: 16 February 2026
URL: https://importai.substack.com/p/import-ai-445-timing-superintelligence
Jack Clark's Import AI issue 445 covers several distinct research threads. Nick Bostrom's 'Optimal Timing for Superintelligence' argues that the benefits of superintelligence justify pursuing it swiftly, with any pause best reserved for a narrow late-stage window. Two new benchmarks are discussed: AIRS-BENCH tests AI agents on contemporary ML research tasks, finding current models below human parity; and First Proof presents frontier unsolved mathematics problems as an ecologically valid creativity test, with leading models unable to solve most questions. Meta's Kunlun recommendation system paper demonstrates scaling laws for ad-targeting AI, with implications for how compute investment in these societally significant systems will grow. The issue also covers an economist's argument that human-touch preferences will sustain human employment even under broad automation.
Implications for Australian agencies:
- [Monitor] Teams tracking frontier AI safety discourse may want to monitor the Bostrom paper and associated debate, as arguments about AGI timing are likely to surface in international policy discussions.
- [Monitor] Agencies interested in AI capability evaluation could note the First Proof and AIRS-BENCH benchmarks as indicators of how the field is approaching autonomous AI research agent assessment.
Retrieved from SIMS, 18 July 2026.