If hundreds of millions of AGI instances are deployed simultaneously by 2027 to automate the algorithms governing their own architectures, the industry will compress a decade of human-led algorithmic progress into less than a year — culminating in 'run...

If hundreds of millions of AGI instances are deployed simultaneously by 2027 to automate the algorithms governing their own architectures, the industry will compress a decade of human-led algorithmic progress into less than a year — culminating in 'runaway superintelligence' via recursive self-improvement. | Announcement of AI system producing AI-architecture-research paper autonomously

Key catalyst: Announcement of AI system producing AI-architecture-research paper autonomously

Watch events: First public recursive-self-improvement demo; frontier-lab automated-ML-research benchmarks

Anthropic, OpenAI, DeepMind all pursuing automated ML research agents 2025-2026; SWE-Bench, RE-Bench demonstrate promise. Actual recursive-self-improvement onset untested.

Beneficiaries (1)

{
  "nia": false,
  "qty": "decade-to-<1yr compression",
  "mode": "FORECAST",
  "role": "Cited-VC/Researcher",
  "context": "Extends AI_008 (Superintelligence 2030 intelligence explosion) with more specific 'hundreds of millions of AGI instances' and 'decade-to-<1-year' compression language. Couples with AI_036 (RLHF fails for ASI), ROB_002.",
  "to_year": 2028,
  "conv_cues": "coined framing; explicit population and compression metrics",
  "direction": "HAPPEN",
  "from_year": 2027,
  "timeframe": "2027-2028",
  "conv_level": "HIGH",
  "milestones": [
    {
      "kind": "llm_pre_event",
      "label": "Aschenbrenner 'Situational Awareness' (165pp) publicly released — anchor thesis",
      "source": "https://situational-awareness.ai/",
      "status": "hit",
      "weight": 0.4,
      "ordinal": -9,
      "source_id": null,
      "confidence": 0.99,
      "source_url": "https://situational-awareness.ai/",
      "expected_date": "2024-06-04",
      "observed_date": "2024-06-04",
      "research_origin": "deep_research",
      "measurement_criterion": "Public release of 'Situational Awareness: The Decade Ahead' essay containing 'hundreds of millions of AGIs' recursive-self-improvement scenario"
    },
    {
      "kind": "llm_pre_event",
      "label": "Frontier model documented to autonomously produce novel ML architecture research",
      "notes": "First documented case of AI doing the AI-research-itself loop is the canonical recursive-self-improvement signal.",
      "source": "https://itcanthink.substack.com/p/how-close-are-we-to-self-improving",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -8,
      "source_id": null,
      "confidence": 0.7,
      "source_url": "https://itcanthink.substack.com/p/how-close-are-we-to-self-improving",
      "expected_date": "2026-12-15",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-06-30",
        "from": "2026-06-01"
      },
      "measurement_criterion": "Peer-reviewed or arXiv ML paper documents AI system autonomously generating novel architecture / training-method improvement that yields measurable benchmark gains"
    },
    {
      "kind": "quartile_checkpoint",
      "label": "Q1 window check-in (25%)",
      "status": "pending",
      "weight": 0.05,
      "ordinal": -7,
      "source_id": null,
      "expected_date": "2027-05-12",
      "observed_date": null
    },
    {
      "kind": "llm_pre_event",
      "label": "Single frontier lab deploys >100M AI-instance compute pool dedicated to research-automation",
      "source": "https://www.dwarkesh.com/p/leopold-aschenbrenner",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -6,
      "source_id": null,
      "confidence": 0.55,
      "source_url": "https://www.dwarkesh.com/p/leopold-aschenbrenner",
      "expected_date": "2027-07-01",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-12-31",
        "from": "2026-12-31"
      },
      "measurement_criterion": "OpenAI / Anthropic / DeepMind / xAI publicly disclose >100M concurrent AI inference instances dedicated specifically to automating algorithmic research"
    },
    {
      "kind": "llm_pre_event",
      "label": "Frontier ML benchmark improvement rate accelerates >2x trailing 5-yr trendline",
      "notes": "Aschenbrenner's '5+ OOMs in <1 year' compressed-progress claim — first detectable acceleration in benchmarks is the canary.",
      "source": "https://situational-awareness.ai/",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -5,
      "source_id": null,
      "confidence": 0.5,
      "source_url": "https://situational-awareness.ai/",
      "expected_date": "2027-07-02",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-12-31",
        "from": "2027-01-01"
      },
      "measurement_criterion": "Stanford AI Index or LMArena documents benchmark capability improvement rate >2x trailing 5
... (truncated)