With quotes, Claude Ai promises more reliable answers

In the Great War of the AI ​​which is currently raging, Anthropic is a slightly less known society of the general public than Google (necessarily) or Openai. However, his conversational agent Claude AI is held in high esteem by its users. And that should not change with the latest anthropic announcement, which launches quotes, a functionality intended to improve the reliability of the responses generated by its Claude models. The purpose of this novelty is to reduce errors by anchoring the responses in specific documents and sources.

It must be said that despite their increasingly impressive efficiency, artificial intelligence models remain criticizable (and criticized) for their tendency to generate incorrect or non -verifiable information. The famous “hallucinations” of AI. To answer this problem, Anthropic has therefore introduced quotes, a new feature accessible via its API and the Green AI platform of Google. This solution allows developers to integrate sources files that Claude 3.5 Sonnet and Claude 3.5 Haiku models then use to provide specific references in their responses.

Rather for developers

According to Anthropic, Quotes will make it possible to considerably improve traceability and especially the verifiability of the information generated. In terms of developers, there will be in theory no longer need to use complex orders or a long prompt engineering process to obtain answers from reliable sources. Developers can import documents in PDF format or raw text into the Anthropic API. Once these files have been integrated, the Claude models analyze the information contained and automatically insert specific references into their responses. These references also refer directly to the exact passages used in the sources documents.

Thus, if a user requires a summary of a report or asks a complex question based on a text corpus, the model will not only provide a detailed response, but also the corresponding quotes. This method is particularly useful in fields such as the automatic summary of documents, complex answers or customer support.

Simplified integration, but not free

Still according to Anthropic, one of the main advantages of quotes lies in its ease of use. Developers can activate this functionality in a few simple steps via the API. A specific command activates the automatic insertion of quotes when processing the documents provided. However, it is important to note that this new feature is not free. Its cost depends on the length and number of sources used. For example, processing a document of around 100 pages costs approximately $ 0.30 with the Claude 3.5 Sonnet model and $ 0.08 with Claude 3.5 Haiku. For developers, this additional cost will therefore be balanced with the benefits in terms of reliability.

The war of the announcements

Finally, impossible to speak of quotes without mentioning the general context in which he intervenes. The main providers of generative AI draw in turn at the start of the year. Besides quotes, Openai has also just unveiled its Operator agentswhile Perplexity launches its Sonor research API and French Mistral AI announced a partnership with AFP. It is interesting to note that all these actors do not only copy themselves and seem to try to trace their own furrow. With quotes, Anthropic is more positioned as an actor focusing on reliability and ethics to stand out.