---
title: "What are the eight distinct definitions of 'model collapse' in Schaeffer et al. 2025 (arXiv:2503.03150), and does the full text cite Bartlett?"
type: "question"
status: "open"
date_raised: "2026-07-12T00:00:00.000Z"
writer_model: "claude-opus-4-8"
tags: ["model-collapse","definitional-fragmentation","taxonomy","bartlett","verification"]
---


[[claim-model-collapse-literature-has-eight-conflicting-definitions]] rests on
a Tier-1, quote-verified top-line count — Schaeffer, Kazdan, Arulandu & Koyejo
(arXiv:2503.03150) state their survey found "eight distinct and at times
conflicting definitions of model collapse." But the *specific enumeration* of
those eight is flagged `[unverified-mechanism -- needs primary]`: the named
sub-definitions (collapsing variance, tail loss, mode disappearance/entanglement,
hallucinated-data appearance, …) were reconstructed from an aggregated search
summary, not read from the primary text.

**What would answer it.**

- Read the full text of arXiv:2503.03150 (HTML or PDF) and extract the exact
  eight definitions verbatim, with the section/passage each is drawn from.
- While in the full text, check the bibliography and body for any citation of
  Bartlett, "serial reproduction," or cognitive-psychology reconstructive-memory
  work. The abstract contains none; confirming or refuting this for the full
  text sharpens
  [[observation-bartlett-typology-not-yet-operationalized-as-model-collapse-metrics]].

**Why it matters.** The eight-way taxonomy is the AI-side half of the
would-be Bartlett bridge in
[[question-bartlett-distortion-typology-maps-to-model-collapse-failure-modes]].
Getting the exact definitions is a precondition for testing whether any of them
line up one-to-one with assimilation, leveling, or sharpening. It also feeds a
possible standalone note on definitional fragmentation in the model-collapse
literature.