RFM RM As User Dataset: 8 Reward Models as Proxies for Human Preferences

A dataset generated for the NeurIPS 2025 paper "Capturing Individual Human Preferences with Reward Features". It uses 8 publicly available reward models as proxies for human raters to support reproducibility in the "Modelling groups of real users" section. The dataset was authored by Google and last updated on 2025-10-29.

Use Cases

• Simulating large-scale human preference data based on the use of 8 reward models as proxies.
• Testing reward model alignment and generalization in group modeling scenarios as described in the paper.
• Benchmarking methods for capturing individual human preferences using reward features.

Strengths

• Dataset is associated with a NeurIPS 2025 paper, indicating academic relevance.
• Uses 8 distinct publicly available reward models to generate proxy data.
• Explicitly released to support the reproducibility of specific experiments.

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 8 source reward models.

Provenance

Source

Google

Collection Method

Generated using 8 publicly available reward models as proxies for human raters.

Time Range

Associated with research for NeurIPS 2025.

Freshness

Last updated 2025-10-29 12:29:15; freshness should be verified.

Geography

null