AI reaches parity with human superforecasters on ForecastBench
On ForecastBench, several AI systems are now statistically indistinguishable from human superforecasters — the top being 'Cassi,' with xAI and Google DeepMind models also at parity. Two honest caveats keep this at verified-but-bounded: the superforecaster comparison was last elicited in 2024, and overlapping 95% CIs mean the data support parity, not clear outperformance. A cheap, tireless, calibrated adviser for insurance, policy, ops, and strategy is arriving.
Why it matters (spine — geo). The geopolitics of compute. Pilot calibrated, Brier-scored forecasting as a decision aid (triage, staffing, budgeting) — and score it against resolution, don't trust it blind.
What would prove it wrong (falsifier). A fresh superforecaster elicitation reopens a durable human lead, or the AI parity fails to replicate out-of-sample.
Live-status check + triangulation. On ForecastBench, several AI systems are now statistically indistinguishable from human superforecasters — the top being 'Cassi,' with xAI and Google DeepMind models also at parity. Two honest caveats keep this at verified-but-bounded: the superforecaster comparison was last elicited in 2024, and overlapping 95% CIs mean the data support parity, not clear outperformance. A cheap, tireless, calibrated adviser for insurance, policy, ops, and strategy is arriving. Source: Analysis · Forecasting Research Institute (+ AIA Forecaster, arXiv 2511.07678) (2025-11-01); see source link. Every post-cutoff figure was tool-fetched against a live source during the Kimi K3 sweep (2026-07-20); see sweeps/2026-07-20-VERIFIED_DIGEST.md for the full provenance and any corrections of record.
Pilot calibrated, Brier-scored forecasting as a decision aid (triage, staffing, budgeting) — and score it against resolution, don't trust it blind.
domains: geopolitics · open-weights · compute