Join our each day and weekly newsletters to receive the latest updates and exclusive content on industry-leading AI coverage. Learn more
OpenAI quietly improved the controls in its Assistants API, making it easier for developers to work with files when building AI assistants.
Improved file search capabilities in Assistant are intended to assist AI agent developers refine their answers. These improvements allow developers adjust how agents select the information from which they generate answers. They can review file search results and tune “the behavior of the file search ranker to change how relevant the results must be before they are used to generate an answer.”
OpenAI launched its Assistants API in November 2023, which it says is a small step toward fully autonomous AI agents.
The Assistants API allows developers to make use of existing OpenAI models with step-by-step instructions when building assistants for different applications. It also allows them to make use of other tools in the OpenAI ecosystem. The Assistants API also enables you to create assistants that may “talk” to other agents.
While OpenAI told VentureBeat when the API first launched that the Assistants API creates assistants that also require guidance — versus agents that perform tasks independently — it’s a step in the right direction for developers trying to build more autonomous AI applications.
It seems that many people have been looking for a option to higher tune their assistants since OpenAI’s updates were met with relief.
Agentic future
AI agents, where AI can routinely perform tasks for users with little or no instruction or prompting, are a major focus for many firms in the AI space. The idea is to take many of the mundane tasks out of the hands of humans, like booking a flight, and let an AI agent or assistant do it for them. For many firms, AI assistants could routinely fill out forms or pull information from a dataset when they see that employees are starting a specific task, like talking to a customer.
Other firms also offer platforms to assist build AI agents. Google recently made its Oscar AI agent platform open source. Meanwhile, firms like Salesforce have begun offering enterprise-specific agents to customers.
AI agents are still relatively recent and still have a lot of room for improvement, helped by features like OpenAI’s file search controls. There’s also no option to fully assess how accurate current AI agents are yet, and some reports have shown that agent benchmarks still miss some metrics.