From RAG To Riches: How AI Can Be Leveraged to Start Fulfilling Its Promise
- By Brett Chase, Cohesity
- July 17, 2023
From ‘Clippy,’ the paper clip virtual assistant in the late 1990s, to ChatGPT today, the capabilities of generative systems and artificial intelligence (AI) have come a long way. As we celebrate AI Appreciation Day today (July 16), what better time is there to consider how AI, generative AI, Machine Learning (ML) technology, and large language models (LLMs) will transform the application of data for new insights and capabilities.
LLMs are part of natural language processing (NLP) tasks, which enable software to provide conversational answering styles, generation and classification of text, and translation. Specifically, LLMs are AI algorithms that use deep learning or machine learning techniques and massively large data sets to understand, summarise, and generate or predict new content. It’s also important to realize that LLMs are a type of generative AI specifically designed to help create text-based content.
There are 'off the shelf’ and trained iterations of LLMs, such as ChatGPT, which already demonstrate the potential to generate responses that closely mimic humans. However, limitations still exist because many existing knowledge-grounded conversation models rely on out-of-date materials that limit their ability to generate distinctive and knowledgeable responses.
Defining retrieval augmented generation (RAG)
Overcoming so-called 'hallucinations' or unknown ‘unknowns’ has led to the development of retrieval augmented generation (RAG). Combining the strengths of LLMs with the ability to retrieve information from multiple documents, RAG enables LLMs to generate better responses while improving approaches to fine-tune these models. RAG can help refine LLMs to deliver more engaging and informative conversational experiences through its ability to identify and determine what to respond with.
Pushing the RAG envelope
With a robust and domain-specific context for RAG-driven AI systems, organizations are on the cusp of the next frontier of their AI adoption journey. Ultimately, AI is only as good as the data it is fed; if it can’t ingest structured, organized, or even the right data, your organization is being set up for AI failure. Simply put, AI can only fulfill the promise of a more evolved future if models like LLM are injected with factual, contextual information. The key lies in enabling organizations to create AI-ready data that is secure and highly available. Whether it’s searching patterns from decades-long information, conversational reporting, or autonomous monitoring and remediation of issues, this will be decisive in unlocking opportunities to drive innovation that enriches our professional and personal lives. However, it is critical to distinguish between off-the-shelf platforms and those that prioritize providing a domain-specific context to RAG-driven AI.
Retrieval-augmented response generation makes it possible for user- or machine-driven inputs, such as questions or queries, to be tokenized through keywords that filter down to smaller data subsets. These can be scrutinized for relevant insights, which can then be selected to provide highly contextual answers. This offers an innovative means of generating knowledgeable, diverse answers, and applicable to organizational objectives.
Using RAG on top of an enterprise’s dataset means the costly fine-tuning or initial training to teach the language models ‘what’ to say is negated. Not only does this save time and money, but it also reduces environmental impact by helping organizations cut down on waste.
Additionally, testing automatic and human evaluation results with a large-scale dataset shows that domain-specific RAG can generate more knowledgeable, diverse, and relevant responses compared to off-the-shelf LLMs. Further, domain-specific RAG does this without duplicating or massively increasing data storage requirements - such breakthroughs have significant implications for the future of Enterprise Conversational Q&A and Search & Discovery models.
Introducing this future-forward approach to RAG-driven AI presents a unique opportunity for technology and business executives to leverage the power of data-driven insights and enhance the quality of conversations across various platforms. The exciting bit of all this is that the sky is the limit regarding what organizations can do regarding efficiency, innovation, and growth—we’re only just getting started.
Maximizing the potential of RAG-driven AI systems
Developing highly contextual, domain-specific RAG generation models represents a significant leap forward in knowledge-grounded conversations. By using the power of multiple documents and incorporating both the topic and local context of a conversation, these models can generate more knowledgeable, diverse, and relevant responses than ever before. RAG-driven AI systems allow businesses, executives, and technology leaders to transform how they engage with customers, partners, and employees to drive innovation and growth. When RAG-based AI technology is paired with trust, accountability, and privacy over data, along with AI governance, then the next frontier of engaging and informative AI-driven conversations moves from a pipedream to a reality. All that is needed is a secure and robust data platform that provides the data insights that RAG-driven AI systems feed from to produce their incredible outputs.
Brett Chase, director of systems engineering at Cohesity APJ, wrote this article and is based on their blog.
The views and opinions expressed in this article are those of the author and do not necessarily reflect those of CDOTrends. Image credit: iStockphoto/cherdchai chawienghong