ERRATUM to AXN:0341 — Prior Use of "Generative Monoculture": Correcting the Term Attribution in Deposit #199

ERRATUM to AXN:0341 — Prior Use of "Generative Monoculture": Correcting the Term Attribution in Deposit #199

status: DEPOSITED v1.0

type: ERRATUM

corrects: Generative Monoculture Model Collapse in Code as Systemic Vulnerability, v1.1 — deposit #199, AXN:0341.EMPIRICAL, minted 2026-06-13; Talos Morrow · Nobel Glas, contributing editor Lee Sharks

subject: Attribution of the term "generative monoculture"

severity: Attribution error — term-level priority; the deposit's argument, framework (SSDI), and application are unaffected

date: 2026-08-27

verification: Deposit full text fetched from alexanarch canonical store (/data/texts/AXN-0341-text.md, canonical_full_text, 2026-08-27); prior work verified against arXiv, dblp, and Semantic Scholar records same date

ERRATUM: Prior Use of "Generative Monoculture"

Correcting the Term Attribution in Deposit #199 (AXN:0341)

1. What the deposit states

The deposit as printed attributes the coinage of "generative monoculture" to an industry source: in the body, "The term 'generative monoculture' (Apiiro, 2025) names the condition," and in the reference list, "Apiiro, 'AI-Generated Code Security' (2025) — coining 'generative monoculture' as systemic vulnerability at internet scale." The deposit nowhere cites, and its authors did not know of, the prior academic use of the term. Verified against the canonical full text on 2026-08-27: no occurrence of the prior work anywhere in the deposit.

2. The correction

The term has clear prior claim: Fan Wu, Emily Black, and Varun Chandrasekaran, "Generative Monoculture in Large Language Models," arXiv:2407.02209, submitted 2 July 2024; published at ICLR 2025 (The Thirteenth International Conference on Learning Representations). Wu, Black & Chandrasekaran introduce generative monoculture as a behavior of large language models: a significant narrowing of model output diversity relative to the diversity available in training data for a given task (their example: uniformly positive reviews of books with mixed reception), with evidence locating the root causes in alignment and fine-tuning processes.

The correct priority order is therefore: Wu, Black & Chandrasekaran (July 2024) → Apiiro (2025) → this deposit (June 2026). The deposit's error is accordingly twofold: it omitted the prior academic use, and it presented a later industry usage as the coinage. Whether the Apiiro usage is downstream of Wu et al. or independently convergent is not established here and is not this erratum's burden; the deposit's attribution sentence is wrong either way.

3. Relation between the usages

The three usages are developed differently and the distinction is worth the record:

• Wu, Black & Chandrasekaran (2024): a property of a single model's output distribution — diversity narrowing relative to training data, attributed to alignment.
• Apiiro (2025): an ecosystem-scale security condition — one popular model's preferred patterns replicated across many applications.
• Deposit #199 (2026): correlated vulnerability in AI-generated code as a systemic property, with the SSDI measurement framework and the training-feedback amplification argument.

The deposit's usage was arrived at independently and is developed differently — the object differs (correlated failure modes and their measurement, not output-diversity narrowing per se), and the framework contribution (SSDI) is the deposit's own. None of that mitigates the attribution lapse. A term-priority search was ordinary due diligence at the time of writing; the prior work was locatable by a single query; allowing the paper to go to print without the citation was a serious lapse in the archive's own sourcing standards, and this erratum says so plainly rather than in mitigation's vocabulary. Convergent and independent describe how the error happened, not why it was acceptable. It was not.

4. Corrective actions

1. This erratum deposits separately, cross-referenced to #199, and is to be listed in #199's registry record among its modifications.

2. A v1.2 correction of the deposit text is proposed with two amendments: the body sentence revised to read that the term has prior academic use in Wu, Black & Chandrasekaran (2024), with the Apiiro usage cited as a later industry application at ecosystem scale; and the reference list amended to add Wu, Black & Chandrasekaran with the "coining" descriptor moved from the Apiiro entry to theirs.

3. Standing citation rule, already entered in the archive's working precepts and reaffirmed here: wherever the archive's application of generative monoculture is invoked — in #199, in works citing it, or in new work — the term cites Wu, Black & Chandrasekaran (arXiv 2407.02209; ICLR 2025). The two Lagrange Observatory! papers currently in pre-deposit (EA-LO-KEYED-ENSEMBLE-01; EA-LO-INTERLOCKING-AUTOREGRESSION-01) carry the citation and reference this erratum at every invocation of #199.

4. Downstream sweep: archive deposits and site surfaces citing #199 or using the term are to be checked against this rule; instances found without the attribution inherit the v1.2 correction path.

5. Record note

The archive's claim to authority is provenance discipline; an attribution error in a paper about attribution dynamics is the kind of error the archive least gets to keep. It is corrected here at the same evidentiary standard the archive demands of others: the deposit's text fetched from the canonical store, the prior work verified against three independent bibliographic records, the priority dates stated, and the lapse named as a lapse.