Photo credit: venturebeat.com
LexisNexis has introduced its AI assistant, Protégé, with the aim of optimizing legal workflows without relying heavily on large AI models. This tool is designed to assist lawyers, paralegals, and associates with drafting and proofreading legal documents while verifying the accuracy of cited materials in briefs and complaints. Rather than providing a general-purpose legal AI assistant, LexisNexis sought to create one that can adapt to a firm’s specific processes and requirements.
According to Jeff Riehl, the Chief Technology Officer of LexisNexis Legal and Professional, the company’s strategy involved harnessing the capabilities of large language models (LLMs) from various providers, such as Anthropic and Mistral, to identify the most effective models for specific user inquiries. Riehl emphasized a tailored approach: “We use the best model for the specific use case as part of our multi-model approach. We aim for the model that offers the quickest response times and optimal results.” This approach sometimes involves using smaller language models, or even distilling LLMs into more compact forms to enhance efficiency and reduce costs.
While LLMs continue to play a significant role in AI development, many organizations, including LexisNexis, are increasingly turning to small language models (SLMs) or distillation techniques. The distillation process, where an LLM imparts its knowledge to a smaller counterpart, is gaining traction among developers seeking efficiency in simpler applications like chatbots and basic code analysis. Protégé was designed to leverage these smaller, agile models for its functionalities.
This is not LexisNexis’s first foray into AI technology. The firm has a history of employing artificial intelligence, especially in natural language processing, deep learning, and machine learning domains. The landscape shifted dramatically with the release of ChatGPT in November 2022, sparking interest in generative and conversational AI capabilities.
Riehl pointed out that LexisNexis’s development involves distinct models from major providers. The LexisNexis + AI platform integrates Claude models from Anthropic alongside OpenAI’s GPT models and capabilities from Mistral, aiding in task-specific solutions. Their multi-model methodology allows for the segmentation of tasks into digestible components, ensuring that each aspect is addressed by the most suitable AI model. For instance, Mistral is utilized for query evaluation initially, followed by models adept at generating or summarizing content.
Currently, Protégé primarily relies on a fine-tuned version of the Mistral model, though LexisNexis has previously employed a fine-tuned Claude model. The company is also evaluating new OpenAI models, particularly those with reinforced fine-tuning capabilities, and exploring the potential for using Google’s Gemini models.
All of LexisNexis’s AI solutions are supported by its knowledge graph, enhancing their retrieval augmented generation (RAG) capabilities. This may pave the way for further advancements in automating legal processes through the PROTÉGÉ system.
Long before the emergence of generative AI, LexisNexis had explored the utilization of chatbots within the legal sector, beginning initial tests in 2017. Now, Protégé is part of the LexisNexis + AI platform, consolidating various AI services provided by the company.
Protégé is crafted to assist law firms with numerous essential tasks typically undertaken by paralegals and associates. This includes drafting legal briefs and complaints, recommending steps in legal workflows, suggesting next search prompts, preparing deposition questions, ensuring accuracy in citations, generating timelines, and summarizing complex legal documents.
“Our vision for Protégé is to serve as a foundational step towards enhanced personalization and agentic capabilities,” Riehl stated. “We envision tailored support for different types of legal professionals—whether they practice in M&A, litigation, or real estate—leading to an increasingly personalized experience based on one’s specific responsibilities.”
As Protégé enters a competitive market, it joins other notable legal technology solutions, including Thomson Reuters’ CoCounsel, which employs a customized OpenAI model, and Harvey, which has garnered significant investment, including backing from LexisNexis.
Source
venturebeat.com