M42 Health, a company based in Abu Dhabi, UAE, has released Med42, an open-access clinical large language model (LLM) that has the potential to revolutionize healthcare. This 70 billion parameter model, fine-tuned from Meta’s Llama-2 – 70B model, outperforms previous open-source medical AI models and achieves up to 72% accuracy in a zero-shot evaluation on the United States Medical Licensing Examination (USMLE).
Med42 was built by the M42 Health AI team using their vast collection of curated medical literature and patient information. The model was fine-tuned using the Condor Galaxy 1 supercomputer and was also assessed by experts at the Mohamed bin Zayed University for Artificial Intelligence (MBZUAI).
Med42 is a free and publicly available LLM that aims to make medical information more accessible. With its adaptability and ability to provide accurate responses to medical inquiries, Med42 has the potential to significantly impact clinical decision-making. It can assist doctors in generating personalized treatment plans based on medical records and accelerate the process of reviewing extensive medical material.
As Med42 becomes available for testing and evaluation, some potential applications include answering health-related questions, summarizing medical history, supporting medical diagnosis, and addressing common health concerns.
The code and weights of Med42 have been released to Hugging Face, encouraging collaboration and further scientific examination. Med42’s licensing terms allow for free research and non-commercial usage while considering the risks and obligations associated with using AI in healthcare.
Key indicators of Med42’s performance include outperforming other publicly available medical LLMs with 72% accuracy on a USMLE sample exam and achieving higher accuracy on the MedQA dataset compared to OpenAI’s GPT-3.5.
However, there are limitations to consider. The therapeutic applications of Med42 are still in the early stages, and extensive human testing is ongoing to ensure safety. There is also a risk of creating misleading or dangerous data and using biased data for training.
Further real-world validation of Med42 is necessary before it can be used in clinical practice. It is crucial to address potential issues such as producing inaccurate or harmful results and addressing existing training data biases. The responsible testing and validation of Med42 are essential to unlock its full potential in improving healthcare decision-making and expanding access to treatment.
The release of Med42 highlights the progress made in medical AI and emphasizes the importance of ethics and safety in its development. The open-access nature of Med42 allows researchers worldwide to benefit from its publication. With thorough validation, models like Med42 can have a significant impact on healthcare decision-making and global access to treatment.
To learn more about Med42 and the research behind it, visit the project page. For the latest AI research news and updates, join the ML SubReddit, Facebook Community, Discord Channel, and subscribe to the Email Newsletter.
If you like our work, you will love our newsletter. Subscribe here.
We are also on WhatsApp. Join our AI Channel on Whatsapp.