Preventing AI Model Collapse through Public Disclosure of Research References AI企業論文参照公示によるモデルコラプス回避
Abstract
This paper proposes that AI companies can prevent model collapse and reduce hallucination by publicly disclosing which papers their models reference or exclude, along with reasons for exclusion. Three contamination sources are identified: spirituality-adjacent papers with circular logic, multi-language mistranslation relay papers, and authority-biased weighting driven by citation cartels. Disclosing non-reference status of specific papers severs contamination propagation at minimal cost. Japanese mistranslation papers are prioritized as the primary hub to cut. Sixteen requirements are defined covering problem definition, principles, strategy, and implementation. AI企業が自社モデルの論文参照情報(参照/非参照とその除外理由)を公示することで、モデルコラプスの回避とハルシネーションの低減が可能であると提案する。汚染源としてスピリチュアル系論文、多言語誤訳リレー論文、引用カルテルに駆動された権威バイアス重みづけの3つを特定した。特定論文の非参照情報を公示するだけで汚染伝播を断てる。日本語翻訳ミス論文をハブとして優先除外する戦略を採用。問題定義・原則・戦略・実装の計16要件を定義した。
Fields of Study & Subjects
Ethics and Social Impacts of AI
Topic Modeling
Academic integrity and plagiarism
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 16 February 2026
- Language
- JA