Organizations of all sizes and sectors can Generative AIThey face a variety of challenges today, from improving operational efficiency to reinventing their business. But as they begin to adopt this transformative technology, they face a common challenge: delivering accurate results.
This is a big problem: bias and other inaccuracies undermine trust, and for generative AI applications, trust is everything.
The solution? Customizing large language models (LLMs), the primary AI technology that powers everything from entry-level chatbots to enterprise-grade AI initiatives.
LLMs alone may provide inaccurate results or results that are too general to be useful. To build true trust among customers and other users of their generative AI applications, companies need to guarantee accurate, up-to-date, personalized responses, which means customizing their LLMs.
However, customizing LLMs is complex, time-consuming, and resource-intensive. It requires specialized knowledge, and not every organization has data scientists or machine language engineers. However, many organizations are opting for proven, cost-effective customization techniques that increase accuracy and relevance while making the most of a resource most organizations already have in abundance: data.
How RAG improves accuracy
Search Augmented Generation (RAG) has emerged as the preferred customization technique for enterprises to rapidly build accurate and reliable generative AI applications. RAG is a fast, easy-to-use approach that helps reduce imprecision (or “hallucinations”) and increase the relevance of answers. It is more cost-effective and requires less expertise than labor-intensive techniques such as fine-tuning LLMs and continuous pre-training.
For generative AI application builders, RAG provides an efficient way to create trustworthy generative AI applications. For customers, employees, and other users of these applications, RAG means more accurate, relevant, and complete responses that build trust with responses that can cite sources for transparency.
Generative AI output is only as good as the quality of the data, so choosing trusted sources is key to improving responses. RAG enhances LLM by retrieving and applying data and insights from organizational data stores or trusted external sources of truth to deliver more accurate results. Even for models trained on stale data, RAG has access to the latest, near real-time information to update the models.
RAG Activities
Food delivery company DoorDash is applying RAG in its generative AI solution to improve self-service and enhance the experience for independent contractors (“Dashers”) who submit high volumes of assistance requests.
DoorDash is working with Amazon Web Services (AWS) to complement its traditional call centers with a voice-activated, self-service contact center solution. At the core of its generative AI solution, DoorDash is using Anthropic’s Claude model and Amazon Bedrock, an AWS service that enables organizations to quickly and easily build and scale generative AI applications.
By using RAG to customize the Claude 3 Haiku model, Bedrock enables DoorDash to access a deep and diverse knowledge base from enterprise sources to provide relevant and accurate responses to Dashers, reducing average response times to under 2.5 seconds. DoorDash’s generative, AI-powered contact center now handles hundreds of thousands of calls every day.
Having access to this vast database through RAG was key to building trust. “We built a solution that ensures Dashers have access to the information they need, when they need it,” says Chaitanya Hari, contact center product lead at DoorDash.
The power of customization
Customization greatly improves the accuracy and relevance of responses, especially for use cases that require the use of up-to-date, real-time data.
RAGs are not the only customization strategy; tweaks and other techniques can play an important role in customizing LLMs and building generative AI applications. But as RAGs evolve and expand their capabilities, they will continue to serve as a way to get started with generative AI quickly and easily, ensure better and more accurate responses, and build trust among employees, partners, and customers.
Learn more AWS-Generated AI.
