Ultrachat Uncensored: Filtered Conversational AI Training Data

Ultrachat-Uncensored is a filtered variant of the Ultrachat conversational dataset. The author branles14 removed examples where bot messages matched specific terms listed in filters.txt, aiming to foster more constructive and neutral conversations. The dataset was last updated on July 20, 2023.

Use Cases

• Training conversational AI models based on filtered dialogue examples.
• Studying the impact of content moderation on chatbot responses.
• Developing neutral dialogue systems based on the described filtering goal.
• Benchmarking language model safety techniques using a pre-filtered corpus.

Strengths

• Explicitly documents a filtering process targeting specific undesirable terms.
• Links to a filters.txt file detailing the removed content.
• Derived from the established Ultrachat dataset, suggesting a known base corpus.

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 is unknown, which may limit suitability assessment.

Provenance

Source

Derived from the Ultrachat dataset.

Collection Method

Created by filtering the original dataset to remove examples where bot messages matched specified terms.

Freshness

Last updated 2023-07-20 03:39:25; freshness should be verified.