Sign in to view source links and access this dataset
Description
788,984 peer reviews for submissions to the ICLR and NeurIPS conferences over an 8-year period. This dataset, created by IntelLabs, pairs human-written reviews with AI-generated counterparts produced by five large language models, including GPT-4o and Claude Sonnet. It was last updated on May 27, 2025.
Use Cases
Training classifiers to detect AI-generated text based on the paired human and AI review structure.
Benchmarking the performance of detection algorithms on a large-scale, domain-specific corpus.
Analyzing stylistic and linguistic differences between human and machine-written peer reviews.
Studying the evolution of AI writing capabilities across different large language models over time.
Strengths
Large scale with 788,984 total reviews.
Covers an 8-year span of submissions from two major AI conferences.
Includes AI-generated text from five widely-used large language models.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Row count for each subset (human vs. AI) is unspecified.
Data may reflect temporal and topical bias inherent to the source conferences.
Provenance
Source
IntelLabs via Hugging Face.
Collection Method
Likely collected from ICLR and NeurIPS conference submission systems, with AI reviews generated using specified LLMs.
Time Range
Covers 8 years of conference submissions (specific years not stated).
Freshness
Last updated 2025-05 27 15:44:37; freshness should be verified.
Geography
null
License is unknown; terms of use must be verified before application.