Summarize From Feedback: Human Feedback Data for Summarization Models

Summarize from Feedback contains the human feedback data released by OpenAI for the 'Learning to summarize from human feedback' paper. The dataset was last updated on January 3, 2023. It is hosted on the Hugging Face platform.

Use Cases

• Training reward models for summarization based on human preference data.
• Fine-tuning language models for summarization via reinforcement learning from human feedback.
• Benchmarking summarization model outputs against human judgments.
• Studying the alignment between model-generated summaries and human preferences.

Strengths

• Data originates from OpenAI's published research on learning from human feedback.
• Dataset was updated on 2023-01 03, providing a specific timestamp.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• 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.

Provenance

Source

OpenAI

Collection Method

Released as part of the 'Learning to summarize from human feedback' research paper.

Time Range

null

Freshness

Last updated 2023-01-03 16:55:41; freshness should be verified.

Geography

null