LexisNexis, an innovator in the field of legal research, recently launched an AI assistant called Protégé, aiming to provide full support to lawyers, assistants and legal specialists. The core features of Protégé include assisting users in writing and proofreading legal documents and ensuring the accuracy of citations in complaints and briefings. LexisNexis' goal is not just to create a universal legal AI assistant, but to hope that Protégé can learn the workflows of various law firms and provide highly customized services.
In the process of developing Protégé, LexisNexis took full advantage of Anthropic and Mistral's advanced large language models (LLMs) and tried to find the models that best meet user needs. "We select the best model based on the specific usage scenario for the best results and fastest response times," said Jeff Riehl, chief technology officer at LexisNexis. In some cases, LexisNexis will choose a small language model (SLM) or distillate a large model to improve performance and reduce costs.
Distillation technology refers to making large models “teach” a smaller model, an approach that is becoming more and more popular in many organizations. Small models usually perform well in applications such as chatbots or simple code completion, which is exactly what LexisNexis wants to achieve in Protégé. In this way, Protégé can reduce resource consumption while remaining efficient.
LexisNexis has extensive experience in AI applications and has explored many times in natural language processing, deep learning and machine learning as early as before launching the LexisNexis + AI Legal Research Center in July 2024. Riehl mentioned that since the launch of ChatGPT in November 2022, the ability of generative and conversational AI has excited them and has promoted the development of Protégé.
When building an AI platform, LexisNexis takes a multi-model approach, using models from different model providers such as Claude, OpenAI's GPT, and Mistral. "We break down each user's tasks into individual components and find the best large language model to support each component." For example, when the user enters a query, Protégé first uses a fine-tuned Mistral model to analyze the query, and then switches to the model that is best suited to complete the task.
Currently, LexisNexis mainly relies on a fine-tuned Mistral model, and in other ways there are plans to use the Claude model. The company is also considering using OpenAI's new model and Google's Gemini model to further enhance its AI platform's capabilities. This multi-model strategy not only improves the flexibility of Protégé, but also ensures its efficient performance in various tasks.
More than just a legal assistant, Protégé can assist in writing legal documents, suggesting the next steps in the legal workflow, generating questions, linking citations to ensure accuracy, generating timelines, and summarizing complex legal documents. Riehl notes that Protégé will be a preliminary attempt at personalized and intelligent legal assistants, who hope that every legal professional will have a personalized assistant to support their specific work needs.
Key points: Protégé is an AI assistant launched by LexisNexis, designed to provide personalized support to legal professionals. LexisNexis uses a variety of large and small language models to improve the response speed and accuracy of AI assistants. Protégé can assist in writing legal documents, generating problems and summarizing complex content, and promote the intelligence of legal work.