M42 LLM answers healthcare questions

  • October 24, 2023
  • Steve Rogerson

UAE-based M42 has unveiled an open-access clinical large language model (LLM) to transform AI across the healthcare sector and create a direct impact on patient care.

The 70 billion parameter, generative artificial intelligence (AI) model provides answers to short and long-form medical questions. The model has been developed to enhance clinical decision-making and increase access to synthesised medical knowledge for healthcare professionals, patients, medical researchers and regulators.

Called Med42, it was trained with the firm’s curated dataset of medical knowledge. When tested on multiple healthcare-relevant datasets, the model outperforms larger closed models, such as ChatGPT 3.5, and achieves a 72% score, in a zero-shot evaluation, on the US Medical Licensing Examination (USMLE) sample exam questions.

Med42 was trained and fine-tuned with Cerebras in collaboration with Core42, a G42 company focused on enabling the delivery of national-scale AI and enterprise AI programmes. A team from Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) evaluated the accuracy of the model.

“Med42 will accelerate global access to healthcare knowledge and embodies our commitment to making a transformative impact across the healthcare sector,” said Hasan Jasem Al Nowais, managing director of M42. “M42 is pushing the boundaries of healthcare innovation and charting a path towards more personalised, patient-centric care. The unique combination of M42’s world-class healthcare expertise and cutting-edge medical technologies allows us to impactfully contribute to the future of AI in healthcare, a future where AI is accessible to all while being rooted in ethics and prioritising the safety of users.”

He said the launch of Med42 firmly established and demonstrated the UAE’s prominent standing in the global generative AI space.

“We are grateful for the support from the Abu Dhabi Department of Health and Abu Dhabi’s wider technology ecosystem,” he said.

Med42 has the potential to act as an AI assistant for healthcare professionals. The capability of the model extends to developing personalised treatment plans by analysing a patient’s medical history to identify the best course of treatment. Other use cases include helping doctors access insights faster and pharmacists decide on correct dosages, and it supports scientists studying treatment options to review literature efficiently.

To promote collaboration and enhancement of the generative AI model, Med42 is available for download on Hugging Face, allowing widespread testing, review and assessment by the scientific and developer community. Licensed on terms similar to Meta’s Llama 2 model, Med42 is free for non-commercial use and research, with controls on use, given the obvious risks that could arise when deploying AI in a healthcare context.

“Innovation isn’t just about leading technologies,” said Ashish Koshy, group chief operating officer of M42. “It is rooted in delivering impactful and safe solutions for real-world challenges. Med42 is not just a tool; it’s a platform for collaboration. M42 aims to catalyse global innovation in AI for healthcare, empowering experts across the globe to refine and expand the model’s applications in various medical domains. Our partnerships with technology leaders, including Cerebras, enable us to bring this vision to life and achieve our long-term transformative objectives.”

The full-parameter fine-tuning of the largest Llama 2 model was accomplished on Cerebras and G42’s Condor Galaxy 1 (CG-1) AI supercomputer. Rapid setup and reduced training time was made possible by the 82Tyte of memory and the 54 million AI cores in the 64 Cerebras CS-2 systems inside CG-1.

“Our strategic partnership with G42 and its portfolio companies is designed to change the AI landscape,” said Andrew Feldman, CEO of Cerebras Systems. “Our collaboration has already delivered Condor Galaxy 1, the largest AI supercomputer in the world, as well as Jais, the leading Arabic LLM, to the community. Today, we helped bring another state-of-the-art LLM to the world, Med42, with huge benefits for the healthcare landscape.”

M42 (www.m42.ae) is an Abu Dhabi-based tech-enabled healthcare company with over 20,000 employees and more than 450 facilities in 25 countries. Established in 2022, following the coming together of G42 Healthcare and Mubadala Health, it combines medical and data-centric technologies to deliver better care.