3,307 values expressed by the Claude AI assistant were extracted and classified from hundreds of thousands of real-world conversations. Anthropic created this dataset using a novel privacy-preserving methodology that did not require human reviewers to access conversation content. The dataset was last updated on April 28, 2025.
Use Cases
- Analyze patterns in AI value expression based on the taxonomy of 3,307 values.
- Study AI behavior 'in the wild' based on data extracted from diverse user interactions.
- Research privacy-preserving data extraction methodologies for conversational AI.
- Investigate the relationship between user tasks and expressed AI values.
Strengths
- Contains a taxonomy of 3,307 distinct values.
- Sourced from hundreds of thousands of real-world conversations, suggesting scale.
- Created with a privacy-preserving methodology that prevented human access to raw content.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- Anthropic
- Collection Method
- Extracted and classified from real-world conversations with the Claude AI assistant using a privacy-preserving methodology.
- Time Range
- null
- Freshness
- Last updated 2025-04 28 17:31:57.
- Geography
- null