Large Language Models for Healthcare, Finance, and Legal: Sampling of Recent Activity

ChatGPT, a large language model (LLM) trained by OpenAI, exploded on the scene as soon as it was launched in November 2022, bringing with it an enormous amount of attention from the public and investors hoping to fund the most innovative startups in the rapidly growing AI space. In March 2023, GPT-4, a much more advanced system was introduced, shining a brighter light on utilizing LLMs for complex disciplines such as, healthcare, legal, and finance. In addition to OpenAI, Google/Deepmind, Anthropic, Baidu, NVIDIA, Microsoft, Hugging Face (and others) have all released LLM models.

LLMs that have been trained on domain specific data are more likely to produce better, higher quality results, especially for “domains like law or medicine [that] contain lots of complex domain knowledge that is not present within a generic pre-training text data set.”

Recent launches/announcements include:

BloombergGPT – Bloomberg has developed BloombergGPT, (announced in March 2023) a new large language model specifically trained on a wide range of financial data, “drawing on the company’s existing data creation, collection, and curation resources.” This model will assist Bloomberg in improving sentiment analysis, named entity recognition, news classification, and question answering.

Hippocratic AI – Hippocratic, according to TechCrunch, emerged from stealth on May 16, 2023 and claims to be the first safety-focused LLM designed specifically for healthcare. Founded by physicians, medical professionals and AI researchers from Johns Hopkins, Stanford, Google and Nvidia, Hippocratic’s AI outperforms leading LLMs including GPT-4 and Anthropic’s Claude on more than 100 healthcare certifications.

Med-PaLM 2 – This LLM from Google Research, is designed to provide high quality answers to medical questions. Med-PaLM 2, introduced in spring 2023 and is under limited testing, has achieved an accuracy of 85.4% on USMLE (US Medical License Exam) questions.

Harvey and Casetext – Harvey, built on top of GPT-4, combines GPT model data with legal-specific data, such as case law and reference materials and then is further trained on a firm’s own work materials. Casetext’s AI legal assistant, CoCounsel, also powered by GPT-4, confirmed in March 2023 that GPT-4 passed the multiple-choice and written portions of the Uniform Bar Exam.

Lexis+AI – LexisNexis announced the launch of Lexis+ AI, which uses content from LexisNexis, in May 2023. This new product uses multiple LLMs “to answer legal research questions, summarize legal issues, and generate drafts of documents such as demand letters or client emails.” Lexis+AI uses the authoritative content from LexisNexis.

Photo by Google DeepMind on Unsplash

Copyright © Copyright 2024 Cottrill Research. Site By Hunter.Marketing