Skip to content
AI

Why do models trained on today's web get progressively worse as AI writes more of it?

82

Möglichkeit

The web is now the primary training corpus for frontier models and is already saturated with AI-generated text that no deployed filter reliably catches. Research published across 2024 to 2026 shows that even a fraction of a percent of synthetic data in a training run triggers distributional collapse over successive generations, narrowing output diversity and degrading tail performance. The feedback loop is structural: models trained this year produce content that contaminates the corpus for next year's training run. Proposed mitigations such as source-level allowlists, watermark filters, and synthetic-data verifiers each have bypass vectors and none has been deployed at web-crawler scale. There is no agreed protocol for identifying and quarantining AI-generated training data before it enters a model.

Warum es wichtig ist

A degrading shared training corpus sets a ceiling on every model built from public data, and that ceiling gets lower with each generation.

Wie ich die Chance bewerte

Der Opportunity Score ist meine persönliche Einschätzung, keine Messung: wie stark es schmerzt, wie oft es auftritt und wie wenig heute existiert, um es zu lösen. Ein höherer Wert bedeutet, dass ich es für lohnender halte, es umzusetzen.

Schweregrad8/10

Wie viel Schmerz es verursacht, wenn es auftritt.

Häufigkeit8/10

Wie oft Menschen tatsächlich darauf stoßen.

Whitespace8/10

Wie wenig gute Werkzeuge dafür heute existieren.

Weitere lösungswürdige Probleme