TADDLE: A Tool-Augmented Agent for Detecting Deficient LLM-Generated Peer Reviews
The article introduces TADDLE, a tool designed to detect deficiencies in peer reviews generated by large language models (LLMs). This system aims to address the challenges of identifying quality issues in LLM-generated reviews, which are often fluent and well-structured. TADDLE utilizes a benchmark of 1,800 reviews and employs a multi-faceted approach to classify and analyze these reviews effectively.
- ▪TADDLE is a Tool-Augmented Agent developed to detect deficiencies in LLM-generated peer reviews.
- ▪The system is based on a benchmark of 1,800 reviews from 50 ICLR 2025 papers, annotated by 18 domain experts.
- ▪TADDLE employs four specialized analysis tools and uses a two-stage semi-supervised learning approach for classification.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.26911 (cs) [Submitted on 26 May 2026] Title:TADDLE: A Tool-Augmented Agent for Detecting Deficient LLM-Generated Peer Reviews Authors:Hanqi Duan, Xiang Li View a PDF of the paper titled TADDLE: A Tool-Augmented Agent for Detecting Deficient LLM-Generated Peer Reviews, by Hanqi Duan and 1 other authors View PDF HTML (experimental) Abstract:LLM-generated peer reviews are increasingly common at major venues, yet their deficiencies are hard to detect because they are uniformly fluent and well-structured. Existing work either classifies authorship without judging quality, or scores quality with features designed for human-written reviews; no prior system detects deficiencies in LLM-generated reviews at the level of individual defect types.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
