Anthropic introduces 'Citations' feature to improve AI reliability
What's the story
Anthropic, a leading artificial intelligence (AI) company, has launched a new 'Citations' feature for its developer API.
The new tool enables developers to anchor Claude's responses in source documents such as emails.
According to the company, the feature improves the precision of its AI models by offering detailed references to "the exact sentences and passages" used to generate responses.
Platform expansion
It is available on Anthropic's API and Google's Vertex AI
The 'Citations' feature is now available not just via Anthropic's API but also on Google's Vertex AI platform.
The expansion is intended to offer developers a more complete toolset for enhancing AI accuracy.
Developers can use source files with the models, which will automatically cite claims inferred from those files, further improving the reliability of AI-generated content.
Use cases
Citations proves beneficial in various applications
Anthropic has also noted a few use cases where the 'Citations' feature can prove especially useful.
These include document summarization, question-and-answer situations, and customer support use cases.
In these cases, the new tool can direct models to cite sources in their responses, making them more accurate and reliable.
Feature restrictions
It is limited to specific models and incurs charges
The 'Citations' feature only works with two of Anthropic's models: Claude 3.5 Sonnet and Claude 3.5 Haiku.
The company also revealed that using this feature could incur charges depending on the length and number of source documents utilized.
For example, a 100-page source document would cost approximately $0.30 with Claude 3.5 Sonnet or $0.08 with Claude 3.5 Haiku, according to Anthropic's standard API pricing structure for Citations use case scenario.