Generative artificial intelligence (AI) models with “agent” capabilities are gaining popularity in the field. These models have the ability to connect with external programs, such as databases or web browsers, to enhance their functionality. OpenAI introduced this concept with its “Assistant” API, aimed at simplifying developers' tasks by providing specific functions for their applications. Anthropic, a competitor of OpenAI, has now released its own version called Tool Use for Claude, which aims to automate tasks, personalize recommendations, and streamline data analysis by integrating AI with external tools and services.
Anthropic originally introduced Tool Use, also known as function calling, when it launched its Claude 3 family of models back in March. Initially available in beta, the company has now made Tool Use generally accessible through its own Anthropic Messages API, Amazon Bedrock service, and Google’s Vertex AI.
So, how does Tool Use work? You input a prompt into Claude, such as “What is the weather in New York?” Claude interprets the prompt and sends an API call to the relevant app, in this case a weather app, to fetch the necessary data. The app then sends the data back to Claude, which generates a natural-language response for you. The choice of which app to call and how to pass parameters is determined by JSON or Python calls formulated by Anthropic’s Language Learning Model (LLM). It’s important to note that the app itself, like the weather app, is not provided by Anthropic; it is the developer’s responsibility to provide the necessary tools.
Anthropic offers three versions of Claude: Haiku, Sonnet, and Opus. Each version has differing levels of sophistication when it comes to formulating requests for tools. Opus, for example, can handle multiple tools simultaneously and is skilled at catching missing arguments compared to the other models. It is more likely to seek clarification in ambiguous cases or when a tool may not be necessary. Haiku, on the other hand, defaults to using tools more frequently even if they may not be directly relevant to the query. It also has the ability to infer missing parameters if they are not explicitly given.
The applications of Tool Use are wide-ranging. StudyFetch, an online learning assistant, utilizes Tool Use to provide students with a more interactive experience in navigating course materials via Claude. Another startup, called Hebbia, uses the technology to automate tasks such as extracting metadata from lengthy documents and streamlining workflows for clients in the financial services sector.
With the release of Tool Use, Anthropic aims to empower developers with the ability to integrate AI seamlessly into their applications, automating tasks and providing personalized recommendations. By combining the power of generative AI with external tools and services, Anthropic is harnessing the potential of AI to enhance efficiency and decision-making in various domains. As more developers explore the possibilities offered by Tool Use, we can expect to see even more innovative applications of this technology in the future.
Use the share button below if you liked it.