Harvey Partners with OpenAI to Develop Custom-Trained Case Law Model for Legal Professionals
-
Recently, Harvey and OpenAI announced a collaboration to develop a custom-trained case law model for legal professionals. This AI system not only possesses complex reasoning capabilities but also handles extensive legal domain knowledge and goes beyond single-model invocation.
It can draft legal documents, answer complex litigation scenario questions, and even identify significant differences among hundreds of contracts. This innovative initiative aims to enhance the efficiency and accuracy of legal work, providing robust technical support for legal professionals.
Harvey was co-founded by Winston Weinberg, a lawyer with antitrust and securities litigation experience, and Gabe Pereira, an AI researcher. They recognized the immense potential of leveraging large language models (LLMs) to synthesize information and present it for lawyer review. In case law research, Harvey team's goal is to create an experience where users can directly copy and paste client questions into the model, which will then provide comprehensive answers and cite all sources.
To achieve this, Harvey team initially tried obvious techniques like fine-tuning base models through public APIs and building Retrieval-Augmented Generation (RAG) systems. However, they quickly realized these techniques couldn't meet the needs of complex, open-ended use cases.
Therefore, Harvey decided to collaborate with OpenAI to build a custom-trained model that could inject new knowledge and reasoning methods about this knowledge into the base model. They started with Delaware case law and eventually expanded to case law across the United States, adding data equivalent to 10 billion tokens to the model. Over the past year, Harvey has emerged as a trusted generative AI platform for legal, tax, and financial professionals. The company's team has grown to over 100 employees, with revenues increasing more than tenfold in 2023. Recently, Harvey secured $80 million in Series B funding from investors including Elad Gil, Kleiner Perkins, OpenAI, and Sequoia, valuing the company at $750 million.
Harvey's case law model boasts several key features. It can handle tasks requiring complex reasoning, which is particularly important for legal professionals. Through customized training, the model has acquired knowledge spanning a wide range of legal domains, enabling it to understand and process various legal queries and tasks.
Additionally, the model assists legal professionals in drafting and reviewing legal documents, significantly improving work efficiency. In complex litigation scenario analysis, the model provides in-depth analysis and solutions. Compared to traditional models, Harvey's customized model delivers more accurate and relevant legal information and solutions, ensuring every sentence has clear source citations. To evaluate the performance of the case law model, Harvey collaborated with ten of the largest law firms for an assessment. They provided lawyers with side-by-side comparisons of outputs from the custom case law model and GPT-4.
The results showed that lawyers preferred the outputs of the custom case law model in 97% of cases. This strong preference was primarily because the custom model provided longer, more comprehensive answers that delved deeper into the details of the issues and covered more relevant case law.
Harvey's next focus is exploring how to combine multiple model calls into a single work output to simplify the user experience. Their vision is to serve as a support team member, assisting associates with complex yet routine tasks, allowing professionals to focus their time on client interactions. In this way, Harvey has significant growth potential not only in the legal field but across all professional service sectors.