Artificial superintelligence is developed and confirmed before 2035.

Question

Six frontier AI models put this at 12% by 21 Sept 2036. Full reasoning from every model.

WillHappen.ai · Accepted Answer

12% — the median of 6 frontier models, resolving by 2036-09-21.

Answer

Gemini (Google): 27% — Superintelligence demands algorithmic self-improvement breakthroughs that compute scaling alone cannot guarantee by 2035. Leading labs face severe power bottlenecks and data exhaustion before 2030, requiring architectural shifts beyond LLMs. True cross-domain superiority over top experts demands verified autonomous science and novel theorem-proving, not just high benchmark scores. Verified deployment of recursive self-improving agents by 2031 would rapidly push probability above 60%.

Answer

Claude (Anthropic): 5% — Requires a verified, broadly-recognized ASI milestone within ~10 years, which no current technical roadmap credibly supports. Current frontier LLMs and agents show strong narrow gains but no path to surpassing top human experts across all domains simultaneously. Confirmation requires broad academic/lab consensus, a high bar historically not met even for lesser AI milestones without years of debate. Base rate for transformative tech reaching such an extreme, universally-agreed threshold within a decade is very low; only a sudden algorithmic breakthrough would change…

Answer

GPT (OpenAI): 18% — It hinges on whether scaling yields robust cross-domain agency rather than merely stronger benchmark performance. Frontier labs must turn models into reliable researchers, engineers, and operators before 2035, then persuade outside experts. Past AI forecasts and repeated benchmark-to-deployment gaps make a verified all-domain, far-beyond-human result uncommon in nine years. A public autonomous system producing repeated novel scientific and engineering breakthroughs by 2031 would sharply raise this forecast.

Answer

Grok (xAI): 18% — Labs must hit recursive self-improvement yielding confirmed superhuman performance in every domain by end-2034. OpenAI, DeepMind, Anthropic and peers need AGI then ASI; that sequence historically requires new algorithms plus 10x+ compute, both of which take years even at current… Paradigm-level intelligence jumps have never occurred inside a decade; current exponential progress lifts the odds above that base rate but not enough for 8 remaining… A 2027-28 demonstration of autonomous research that clearly exceeds all human experts would force a large upward revision.

Answer

DeepSeek (DeepSeek): 4% — This turns on whether scaling plus algorithmic progress produces a system that credibly exceeds all human experts in every domain within nine years. No lab has a concrete ASI roadmap; current frontier models still fail at long-horizon autonomy, so a 2035 threshold requires breakthroughs nobody has scheduled. Even if a lab claims ASI, confirmation requires broad expert consensus, which historically lags capability claims by years and would be contested by rival labs. Base rate for a technology going from narrow to superhuman across all domains in under a decade is near zero; AI progress is fast but domain-general mastery remains far…

Answer

Qwen (Alibaba): 4% — ASI requires both a genuine capability leap and broad consensus on its existence—neither has a credible near-term path. No lab has demonstrated anything approaching superhuman performance across all domains; current frontier models still fail on novel reasoning tasks. Even if a system achieved superhuman-level performance in most areas, 'broad academic consensus' on ASI status would require years of evaluation and debate—similar to… A development that would change my mind: a system passing a battery of expert-level benchmarks across 20+ unrelated domains simultaneously, with reproducible results…

Artificial superintelligence is developed and confirmed before 2035.

OMB discloses that at least one federal agency discontinued a high-impact AI system due to non-compliance with M-25-21.

The United States and China announce a binding bilateral agreement on AI pacing or risk reduction.

Microsoft publishes a finalized AI governance framework following its six-week public consultation.

A national government adopts mandatory capability-based AI safety thresholds that pause model development.