Generative AI is here to stay, and the industry recognizes it. In a recent survey of lawyers, 82 percent said they believe generative AI can be applied to legal work and 51 percent said it should be.
Of course, organizations are still appropriately cautious about this new technology, especially when it comes to the privacy and security of client data and the accuracy of results—but more and more are taking a proactive approach to evaluating these new tools. Their question isn’t “What are you doing with generative AI?” but rather “What are you doing with generative AI to make sure you meet the needs of today and tomorrow?”
Making a smart and effective AI investment means understanding how a solution adds value now, but also how it will improve in the future alongside rapid changes to the underlying technology.
At Relativity, we understand the importance of keeping up with the pace of AI advancement. From robust testing of new foundational models, to combining generative AI with other machine learning components, to delivering product enhancements based on customer feedback, we are always exploring ways to quickly, and responsibly, enhance our solutions so customers can drive more value for their businesses and clients. Read on to learn how the continuous innovation that comes with Relativity aiR, our generative AI solutions, will set you up for success no matter what new AI developments the future may bring.
New Models Mean More Testing and Better Results
It’s hard to stay on top of all the generative AI technologies available for use today. In just the last year we’ve seen not only the introduction of large language models with remarkable capabilities, but their continued expansion into new forms of input and output, with the most recent multimodal large language models (MLLMs) able to leverage images, speech, and text. Microsoft Azure’s OpenAI service currently has over a dozen models available, each with different capabilities and many versions that reflect the latest updates. And then of course there are other providers, like Anthropic, Google, and Meta, with their own generative AI models—adding even more options to choose from.
It can be challenging to navigate this complex playing field, but Relativity is committed to testing what’s available and confirming that aiR’s solutions leverage the best underlying technology for their distinct use cases, which, to date, have been Azure OpenAI’s GPT models. When Azure OpenAI’s GPT-4 Turbo (GPT-4T) was released for limited preview in late 2023, Relativity’s applied science team went straight to work, meticulously evaluating each of GPT-4T previewed releases. The testing showed that a transition to this new underlying model would deliver more powerful and precise results, faster.
“The underlying model is one component of our AI solutions, which are complex and nuanced systems. There are interactions between case information, ambiguous language, other types of AI technology, the wide variety of documents in review, and the large language model itself. We not only directly tested the underlying large language model but also the full system, evaluating its performance quantitatively and qualitatively. We fully expect to repeat this process as innovation continues in this space and new models are introduced,” says Aron Ahmadia, senior director of applied science at Relativity.
GPT-4T became fully available for production through Azure in May 2024, and by June, it was implemented to help aiR support projects of up to 100 million documents and analyze thousands of documents per hour across each matter. With nascent models like GPT-4 Omni on the horizon, Relativity’s data scientists are back to work to determine if newer models can help deliver even more efficient document review—and if so, implement it as quickly as possible.
A Multifaceted AI Approach to Solve Complex Privilege Problems
Relativity aiR for Privilege launched with limited availability in April 2024 to help RelativityOne customers accelerate privilege review and privilege log creation. But just a few months prior, the development team was working to make a fast, yet essential, decision on how exactly to incorporate new generative AI technology into the solution.
Early versions of Relativity aiR for Privilege leveraged advanced machine learning algorithms and social network graphs to identify privilege-conferring individuals and legal advice or content being shared to identify potentially privilege documents. While the product delivered strong results, the new wave of generative AI technology introduced even more potential. Relativity Director of Product Management Peter Haller recognized the importance of expediting experiments with customers to test different options and determine the best path forward when it came to leveraging generative AI.
“We quickly explored different ways to use generative AI in aiR for Privilege. First, we tested the product using GPT models on their own, allowing clients to outline their own definition of privilege per matter. Then we tested the use of GPT models in conjunction with the other large language models, machine learning technology, and social network graphs that were already implemented in the initial solution. What we found was that our unique combination of AI technologies was far superior at understanding the nuance of privilege than if generative AI was used by itself and removed the need for client prompt engineering,” Peter explains.
Through these experiments the team determined that a multifaceted AI approach, one that strategically incorporated generative AI alongside other AI capabilities, had tremendous benefit. It reduced upfront setup time, delivered extremely high recall, increased precision across predictions, and provided enhanced reasonings and citations to support decisions. In just two months the team conducted experiments, made the decision to incorporate generative AI, and revamped development plans to support these changes in the limited release program.
Nonstop Investment and Innovation
Each day, Relativity’s dedicated teams of product managers, designers, data scientists, and engineers work diligently to further evaluate each aiR solution, gather feedback, and provide recommendations on how to incorporate new AI advancements.
aiR’s current limited availability programs help these teams collaborate with customers to determine what enhancements should be prioritized. For example, with prompt writing being such an essential component in using aiR for Review, Relativity worked closely with customers to develop a tailored prompt criteria application. This enhancement, which will deliver a more seamless experience for crafting and iterating on prompt criteria, will be available with aiR for Review’s full release later this year.
And things are just getting started. AI advancement isn’t slowing down any time soon, and neither is Relativity’s investment in it. You can be confident that aiR will keep up with an ever-increasing pace of AI transformation—so you and your clients can, too.
Want to learn more about our newest aiR solution, aiR for Privilege? Sign up for our upcoming AI Advantage Webinar to hear from current limited general availability customers who are using the solution and transforming privilege review with generative AI.Sarah Green is a product marketing manager at Relativity, leading go-to-marketing strategies for the company’s newest innovations and partnering closely with cross-functional teams on product launches and adoption activities.
