LiveAI model qualityWestern cycleAnchors pending

By end of 2029, a major AI lab will publicly admit its model got worse because training data was contaminated by AI output?

As of 2 Oct · 12:00 UTCResolves 31 Dec 20293 movers1 pressure moments
Timeline A · Admitted PubliclyReality
50%
Crowd 55%Gap 0

By late 2029, a major AI lab—OpenAI, Anthropic, Google DeepMind, or Meta—releases a technical report or public statement acknowledging that one of its recent models showed measurable performance drops in specific tasks because its training dataset absorbed too much AI-generated text. The lab describes the contamination source (synthetic data from competitors, autonomous systems, or internet scraping) and quantifies the degradation (e.g., reasoning accuracy down 3–8% on benchmark tests). This admission comes after internal audits, user complaints, or competitive pressure expose the problem.

Timeline B · Never AdmittedReality
50%
Crowd 45%Gap 0

Through 2029, no major AI lab makes a public acknowledgment linking model quality loss to AI-output contamination of training data. Labs either avoid the problem through data filtering, train on older pre-2020 corpora, or see performance gaps small enough not to trigger disclosure. Any model degradation gets attributed to other causes—architectural trade-offs, scaling limits, or benchmark saturation—rather than data poisoning.

crowd 55
The quiet paththe one nobody’s watching

Geopolitical or safety crisis overshadows internal technical admissions about model quality.

6%
The read
Reality allows both Admitted Publicly and Never Admitted.

Reality sits near a coin flip between Admitted Publicly and Never Admitted — neither side has decisive weight right now. The main thing that could shift this is Share of web content that is AI-generated.

Where reality is heading

A, 80% bandCrowd today
Projected reality support to 31 Dec 2028A rises from 41% to about 52% with an 80% band of 33 to 71. B falls from 53% to about 44%. Quiet path near 4 to 6%. Crowd today: A 55%.100%7550250Crowd today · A 55%A ~52%B ~44%P ~4%NowJan 2027Jan 202831 Dec 2028
Reality crosses toward A in early 2028. Today’s crowd number sits at the top of A’s band, possible but optimistic. View as table

The movers, ranked by leverage

01
Collective pull shiftMarket-class: discontinuous move in collective attention + capital toward an outcome, measurable in prediction markets (Polymarket / Kalshi / Manifold / Metaculus / PredictIt) and sector-rotation flows. Per AMENDMENT 47 §MCA-2, this IS the B_leak = consciousness measurement instrument where it covers an outcome. Direction contextual depending on which outcome the pull targets.
leverage 50
Prediction markets reprice when collective pull on an outcome moves.
02
Institutional trust decaySignal-class: measurable datum tracking collective trust in institutions — Pew/Gallup polling on Congress/Supreme-Court/media/science/business, approval ratings, cross-party cooperation rate, voter-turnout participation vs grievance, protest-event counts. Direction typically Timeline-B on democratic-haywire cycle.
leverage 55
Falling institutional trust accelerates the destabilization timeline.
03
Institutional defenderActor-class whose role is preservation of existing institutional architecture against erosion pressure. Courts, independent agencies, free-press leadership, academic institutions, professional-standard bodies. Direction typically Timeline-A on democratic-haywire cycle.
leverage 60
Defenders of courts, press, elections keep institutional timelines on track.

What would change our mind

Signals toward A
  • Share of web content that is AI-generated readings confirming the Admitted Publicly direction
  • Lab competitive positioning trajectory bending toward Admitted Publicly
  • Independent evaluation findings moving in support of Admitted Publicly
  • Fresh readings from Shumailov et al. Nature (2024-07-24) — Original model-collapse paper pointing toward Admitted Publicly
Signals toward B
  • Share of web content that is AI-generated readings confirming the Never Admitted direction
  • Lab competitive positioning trajectory bending toward Never Admitted
  • Independent evaluation findings moving in support of Never Admitted
  • Fresh readings from Graphite (Forbes) (2025-10-01) — 57.1% of new online written content likely AI-generated pointing toward Never Admitted

Pressure moments

AI capability discontinuityfrontier model benchmark saturation + new benchmark emergence · autonomous-agent task-completion rate on real-world browser/code benchmarks · model-collapse regression indicators
3-9 months at observable-step window; prelude years
High

Evidence