In a strategic move that seems to coincide with OpenAI’s latest announcement, Anthropic has introduced an exciting new addition to its developer toolkit: Citations. This feature is designed to enhance the reliability of responses generated by the AI models from the Claude family by grounding them in source documents like emails.
Citations allows AI models to provide precise references, pointing to the specific sentences and passages from documents that inform their responses. This feature is now accessible through Anthropic’s API as well as Google’s Vertex AI platform, offering developers a seamless integration.
According to Anthropic, Citations can significantly benefit applications such as document summarization, question and answer sessions, and customer support by ensuring that models cite the origins of their claims. This capability can greatly improve the trustworthiness and accuracy of AI interactions.
However, it’s important to note that Citations is limited to specific models, namely Claude 3.5 Sonnet and Claude 3.5 Haiku. Additionally, the feature comes at a cost. Charges may apply depending on the size and number of source documents. According to Anthropic’s pricing structure, using Citations with a document spanning approximately 100 pages could cost around $0.30 with Claude 3.5 Sonnet, or $0.08 with Claude 3.5 Haiku. For developers aiming to reduce AI-induced errors and hallucinations, this investment may well pay off in terms of enhanced performance and accuracy.
Citations is poised to provide developers with a powerful tool to ensure that AI-generated content is both accountable and anchored in verified information, paving the way for more reliable AI applications across various industries.
