Join our daily and weekly newsletters for the latest updates and exclusive content in the industry’s leading AI coverage. Learn more
When a legal research company Lexisnexis AI Assistant Protégé created, wanted to understand the best way to use his experience without placing a big model.
Protegé aims to help lawyers, Associates and Paralegals to help write and prove legal documents and have something they refer to Complaints and brief information are accurate. However, LexisNexis did not want a general legal AI assistant; They wanted to build something that learns a firm’s workflow and more customizable.
Bring the power of the LexisNexis, large language models (LLS) the power of the anthropic and the Mistral (LLMS) and the user has the best models to find the best models that meet the CTO of Jeff Riehl, LexisNexis.
« We use the best model for specific use of our model approach. We use the model that provides the best result with the fastest response period, » Rehl. « For some use, it will be a small language model like the Mistral or conduct an expulsion to improve performance and reduce costs. »
Although the LLMS still values in AI applications, some organizations take turns to use LLMS to become smaller language models (SLMS) or smaller versions of the same model.
Dismanship where a LLM is a smaller model « teaches » popularize for many organizations.
Small models often work best for applications such as LexisNexis’s Protegé, like Chatbots or simple code completion.
This is not the first time in LexisNexis AI applications before launching LexisNexis + AI in July 2024.
« We’ve used a lot of AI, natural tongue processing in the past, some deep learning and machine learning, » said Rehl. « This was very much before the stage of AI, many AI opportunities were behind the stage, because it was launched in November 2022. But once the chaeatgpt came out, it was very interesting for us. »
Small, subtle adjustable models and model routing
Rehl, LexisNexis, while setting up AI platforms, said that most of the main model providers use different models. LexisNexis + AI used Claude models of anthropic, Openai’s GPT models and a model of the autrral.
This multimodal approach has broken down every task user who wants to perform on the platform. To do this, he had to archit the LexisNexis platform Switch between models.
« We will split any task made to individual components and then determine the best large language model to support this component. We will use the step to evaluate the request of the user in order, » said Rehl.
The company for Protegé, faster response times and models became more delicate for legal use. Thus, Riehl calls the « subtle adjustable » versions, LLS or distilled models in essence, the smaller weighing version.
« You don’t need a GPT-4O to make an estimate of a survey, so we use it for more complex jobs and remove the models. »
A user is a subtle mistral to estimate the query, to assess the query, to assess the question and intention of the survey, « The pings of the first model are a subtle mistral to evaluate the query, to assess the query. Riehl, the search engine or the result of the search engine or result He said it could be an LLM that creates new surveys for another model.
Currently, LexisNexis, when Riehl first comes out, we do not use it today. LexisNexis is also interested in using other Openai models, especially since the company has released Strengthening fine adjustment capabilities last year. LexisNexis is in the process of evaluating Openai’s justification models, including O3 for platforms.
Riehl added that it can also use Google from Gemini models.
LexisNexis can help start the AI platforms all AI platforms with their knowledge schedule, especially Protegé Agency processes.
AI Legal Suite
In the face of the arrival of a Generative AI, Lexisnexis tried to work in the legal industry of Chatbots. In 2017 The company tested the AI Assistant This sits on the company’s Lexisnexis + AI platform, which brings together IBM Watson with Watson.
Protégé helps law firms with prone behavior of paraleegal or common partners. Helps legal brief information and complaints based on documents and data, questions for the next steps, changes and discoveries, new proposals and of course to create progresses, prepare evadels, and of course, to create complex legal documents.
« We see Protégé as an initial step in personalization and agent, » said Rehl. « Think about various lawyers: M & A, Litigators, Real Estate. This will continue to be more personalized based on the specific task you do.
Protégé is now fighting against other legal research and technology platforms. Thomson Reuters takes over Openai’s O1 Mini-model model Cocounsel Law Assistant. Harvey, this Won $ 300 million Investors, including LexisNexis, also have a Legal AI Assistant.
Source link
Leave a Reply