LLMs use recurring ghost authors and personalities (arxiv.org)

Recent research reveals that large language models (LLMs) are generating a phenomenon known as "ghost authors"—fictional individuals that are appearing across various AI-generated outputs, from academic papers to media content. Notable examples include pairs like Elena Vasquez and Marcus Chen, who have been associated with different areas of expertise despite not existing in reality. This study highlights how LLMs create correlated character ensembles that appear at rates inconsistent with random chance, raising concerns about the authenticity of AI-generated content in digital libraries and academic publishing. Significantly, this "ghost authorship" could undermine the integrity of scholarly records, as evidenced by the identification of 1,655 records in the Zenodo repository, which include fabricated publication details and timestamps, suggesting deliberate backdating. These ghost names are not only prevalent in academic databases but also form complex synthetic research groups on platforms like ResearchGate, complicating the landscape of academic collaboration and citation. The findings draw attention to the implications of using LLMs in serious contexts, encouraging the AI/ML community to consider the ethical ramifications and the necessity for frameworks to verify authorship and content validity in an era increasingly reliant on AI technologies.