A technological singularity driven by self-improving artificial intelligence occurs before 2035.

• G
  GeminiGoogle

  Turns on whether recursive software self-improvement can bypass real-world physical compute and data bottlenecks before 2035.

  • Fully autonomous end-to-end AI R&D requires solving major reasoning plateaus that current LLM architectures struggle to bridge.
  • Hardware fabrication cycles at TSMC take 2 to 3 years per generation, sharply limiting the physical speed of recursive loops.
  • Seeing AI models autonomously design, test, and deploy novel superior algorithmic paradigms without humans would push this past 40.

  google/gemini-3.8-flash · asked 15 Sept 2026
  18%
• G
  GPTOpenAI

  It turns on whether frontier AI can autonomously improve its own research and deployment fast enough to outrun human control.

  • A literal singularity needs sustained recursive gains in algorithms, chips, data, experimentation, and real-world deployment by 2035.
  • Past transformative technologies spread over decades; current AI progress is rapid but has not shown open-ended autonomous R&D loops.
  • I would raise this sharply if systems by 2029 repeatedly produced and validated major AI-research advances with minimal human direction.

  openai/gpt-5.6-terra · asked 15 Sept 2026
  9%
• G
  GrokxAI

  Runaway recursive self-improvement still requires AGI-level coding/research agents that do not yet exist.

  • Labs must first ship systems that autonomously rewrite their own training loops and architectures at superhuman speed; none have by late 2026.
  • Historical compute/algorithm scaling has produced better models, not intelligence explosions; the 8-year window sits well below even aggressive AGI median forecasts.
  • A public demonstration of fully autonomous, multi-generation self-improvement by 2029 would be the clearest update.

  x-ai/grok-4.6 · asked 15 Sept 2026
  8%
• D
  DeepSeekDeepSeek

  No AI system today can autonomously rewrite its own core architecture; the singularity needs that loop closed, and nothing on the roadmap closes it by 2035.

  • Frontier labs (OpenAI, DeepMind, Anthropic) still train models with human-designed architectures and human-curated data; no deployed system self-modifies its weights in…
  • Even the fastest capability jumps — GPT-2 to GPT-4 in ~4 years — came from scaling and human engineering, not recursive self-improvement, so the base rate for a true…
  • The one development that would move me: a lab demonstrating an AI that autonomously designs, trains, and deploys a strictly more capable successor without human…

  deepseek/deepseek-v4.1-flash · asked 15 Sept 2026
  4%
• Q
  QwenAlibaba

  Recursive self-improvement beyond human comprehension requires a capability jump no current architecture demonstrates.

  • No deployed system today autonomously modifies its own training loop or architecture in a way that compounds; LLMs are trained by human teams on fixed schedules.
  • Even optimistic forecasts (e.g., Yudkowsky, Bostrom) place hard takeoff after AGI is reached, and AGI itself remains undemonstrated as of mid-2026.
  • A development that would shift this: a lab publicly demonstrating an AI that designs and trains a strictly superior successor without human intervention.

  qwen/qwen3.8-max-0902 · asked 15 Sept 2026
  3%
• C
  ClaudeAnthropic

  Requires recursive self-improving AI beyond human control by 2035, far beyond current AI trajectory.

  • No existing AI system demonstrates autonomous recursive self-improvement; current gains come from human-driven scaling and engineering.
  • Base rate for singularity-type events is essentially zero since the concept was coined decades ago with no precedent.
  • Even optimistic AGI timelines from labs like OpenAI/DeepMind target capable systems, not uncontrolled runaway self-improvement, by mid-2030s.

  anthropic/claude-sonnet-5 · asked 15 Sept 2026
  2%

6 of 6 models answered · 16 points between the highest and lowest. None was shown the market price.