Tracing AI Plagiarism: AI-Generated Books Contain Over 40% Distinctive Phrases
allen_ai · x · 2026-08-01
AI2 leveraged the infini-gram engine to analyze the linguistic origins of AI-generated text, revealing heavy reuse of distinctive expressions from existing corpora.
- Methodology: By counting "distinctive phrases" that appear in five or fewer books on Google Books and nowhere on the web, researchers distinguished genuine original writing from AI-generated stock lines.
- Data Comparison: Among top-selling self-published Amazon books flagged with substantial AI text, distinctive phrases accounted for 41.6% of the text. Unflagged books stood at 37.2%, while award-winning books contained only 19.1%.
- Case Study: Analyzing the prize-winning story at the center of the "GrantaGate" controversy, researchers found its distinctive fragments heavily matched existing content on a fan fiction site.
Related event: Ai2's infini-gram Engine Tracks AI Writing Plagiarism(9 posts)→
More from Research
- Humanoid Robot Dodges 19/20 Thrown Balls Using Onboard Sensors — ChongZzZhang · 2026-08-01
- TMLR Adopts Fractional Authorship, Weighing Credit by 1/k per Author — thegautamkamath · 2026-08-01
- ACE-Data-0: A Large-Scale Multimodal Dataset for Embodied AI — liuziwei7 · 2026-08-01
- Waterloo's R2L Lab to Recruit PhDs, Focusing on Agents and Reasoning Research — hllo_wrld · 2026-08-01
- AgentIR: Deep Research Agents That Leverage Reasoning Context for Retrieval — hllo_wrld · 2026-08-01
- Looped Model Architecture: 8B Params Outperform 32B in Reasoning — SonglinYang4 · 2026-08-01