Post-AGI (2027+), a decade of human-led algorithmic progress will be compressed into ~1 year or less as AGIs automate AI research.

Post-AGI (2027+), a decade of human-led algorithmic progress will be compressed into ~1 year or less as AGIs automate AI research. | Post-AGI capability jump rate

Key catalyst: Post-AGI capability jump rate

Watch events: AI-automated ML research outputs; published-paper authorship tracking; frontier model capability jump rates.

Recursive self-improvement (Wissner-Gross SEM-adj call: 'already here, not 10 years out') showing early signs via AlphaEvolve; magnitude uncertain.

{
  "nia": false,
  "qty": "decade compressed to <1yr",
  "mode": "PROPHECY",
  "role": "Cited-Researcher",
  "caveats": "Requires AGI to be achievable in the first place; requires alignment to not block scaling.",
  "context": "Core 'Intelligence Explosion' thesis: hundreds of millions of instantiated AGIs working 24/7 on neural-architecture optimization at electronic speeds.",
  "to_year": 2029,
  "conv_cues": "recursive self-improvement; explicit compression",
  "direction": "NUMERIC_TARGET",
  "from_year": 2027,
  "timeframe": "post-2027",
  "conv_level": "HIGH",
  "milestones": [
    {
      "kind": "llm_pre_event",
      "label": "Frontier model scores top-tier on automated AI R&D benchmark (METR/RE-Bench)",
      "notes": "Pre-AGI signal — model needs to do AI research before automating it.",
      "source": "METR (Model Evaluation and Threat Research), Anthropic/OpenAI evals",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -8,
      "source_id": null,
      "confidence": 0.65,
      "expected_date": "2026-12-15",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-06-30",
        "from": "2026-06-01"
      },
      "measurement_criterion": "Frontier LLM achieves human-expert parity (≥80%) on METR RE-Bench or equivalent automated AI research benchmark"
    },
    {
      "kind": "llm_pre_event",
      "label": "Major lab claims '≥30% of internal AI research time saved by AI agents'",
      "notes": "Aschenbrenner's RSI thesis — labs already report meaningful R&D acceleration.",
      "source": "Lab blog posts, earnings calls, conference talks",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -7,
      "source_id": null,
      "confidence": 0.7,
      "expected_date": "2027-05-02",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2027-12-31",
        "from": "2026-09-01"
      },
      "measurement_criterion": "OpenAI, Anthropic, DeepMind, or xAI publicly claims ≥30% of researcher-hours augmented or replaced by AI agents in internal R&D"
    },
    {
      "kind": "quartile_checkpoint",
      "label": "Q1 window check-in (25%)",
      "status": "pending",
      "weight": 0.05,
      "ordinal": -6,
      "source_id": null,
      "expected_date": "2027-07-01",
      "observed_date": null
    },
    {
      "kind": "quartile_checkpoint",
      "label": "Q2 window check-in (50%)",
      "status": "pending",
      "weight": 0.05,
      "ordinal": -5,
      "source_id": null,
      "expected_date": "2027-12-29",
      "observed_date": null
    },
    {
      "kind": "llm_pre_event",
      "label": "Algorithmic-progress effective-compute doubling time falls below 6 months",
      "source": "https://aiprospects.substack.com/p/the-reality-of-recursive-improvement, Epoch AI",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -4,
      "source_id": null,
      "confidence": 0.45,
      "source_url": "https://aiprospects.substack.com/p/the-reality-of-recursive-improvement",
      "expected_date": "2028-01-01",
      "research_origin": "deep_research",
      "expected_date_range": {
        "to": "2028-12-31",
        "from": "2027-01-01"
      },
      "measurement_criterion": "Epoch AI / equivalent measures effective-compute doubling time below 6 months (vs ~6-12 months baseline) attributable to algorithmic gains"
    },
    {
      "kind": "llm_pre_event",
      "label": "AGI declaration by ≥1 frontier lab",
      "notes": "Pre-condition for the claim — '10 years compressed into 1 year' requires AGI first.",
      "source": "Lab announcements, capability evaluations",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -3,
      "source_id": null,
      "confidence": 0.3,
      "expected_date": "2028-01-01",
      "research_origin": "training",
      "expected_date_range": {
        "to": "2028-12-31",
        "from": "2027-01-01"
      },
      "measurement_criterion": "OpenAI, Anthropic, 
... (truncated)