PaperWrite-Bench: Evaluating AI-Written Papers for Presentation and Hallucination

PaperWrite-Bench is a dataset for evaluating presentation and hallucination in AI-written papers, created by researchers at The University of Tokyo. The dataset is hosted on Hugging Face and was last updated on 2026-04 02 09:19:29. It is designed for rigorous evaluation of AI-driven research automation and its associated risks.

Use Cases

• Benchmarking AI text generation systems for factual accuracy based on the described hallucination detection task.
• Evaluating the presentation quality of AI-generated research papers based on the described evaluation framework.
• Training or fine-tuning models for research automation tasks based on the dataset's focus on AI-written papers.
• Studying the risks and limitations of coding agents in academic writing based on the described background.

Strengths

• Dataset is hosted on a major platform (Hugging Face) with associated project page, paper, and code.
• Created by researchers from The University of Tokyo, suggesting academic rigor.
• Last updated on 2026-04-02 09:19:29, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific sources and methods used for its creation.

Provenance

Source

The University of Tokyo (hal-utokyo)

Collection Method

Likely created for academic research on AI evaluation, as described in the background.

Time Range

null

Freshness

Last updated 2026-04-02 09:19:29.

Geography

null