limitgen-eval
Can LLMs Identify Critical Limitations within Scientific Research? A Systematic Evaluation on AI Research Papers — Xu et al. (2025) (arXiv:2507.02694, 2025)
What this evaluates
Evaluates whether LLMs can accurately identify and articulate critical limitations in scientific research papers across methodological, experimental, analytical, and literature-related dimensions. The benchmark probes the model's ability to ground critiques in domain-specific best practices and produce actionable, substantive feedback rather than superficial presentation critiques.
Datasets
- LimitGen — total 2000; splits: synthetic (1000), human (1000); repo https://github.com/yale-nlp/LimitGen
Metrics
Limitation Quality(primary) — range: [0, 1]- Not explicitly defined in the provided section. The task requires generating a limitation description for a given paper and aspect, evaluated against ground truth references.
Input / output format
Input: A scientific paper (text or JSON) and a specific aspect of limitations (e.g., Methodology, Experimental Design, Result Analysis, Literature Review).
Output: A generated limitation description reflecting the paper's quality with respect to the specified aspect.
Scoring recipe
# Pseudo-code for evaluation (metric not explicitly specified in text)
def evaluate(predictions, gold):
# Likely uses LLM-as-a-judge or human annotation
# Compare generated limitation text against ground truth description
# Score based on semantic similarity, factual correctness, and actionability
score = judge_model(predictions, gold)
return score
Common pitfalls
- Confusing superficial presentation or style critiques with substantive methodological or analytical limitations.
- Failing to ground identified limitations in established scientific principles or domain-specific best practices.
- Omitting actionable suggestions or concrete steps for improvement, which the taxonomy requires for meaningful limitations.
Evidence (verbatim from paper)
The LLM is tasked with generating the limitation for the given paper, reflecting its quality with respect to the specified aspect.
Citation
@misc{xu2025limitgen,
title={Can LLMs Identify Critical Limitations within Scientific Research? A Systematic Evaluation on AI Research Papers},
author={Xu et al. (2025)},
year={2025},
note={arXiv:2507.02694}
}
- arXiv: 2507.02694