In order to investigate safe and beneficial use scenarios, Google Cloud has launched a restricted test of Med-PaLM 2, a medical language model, for a select group of clients. Med-PaLM 2 is able to lead educational discussions, respond to challenging medical inquiries, and glean insights from unstructured medical materials. It may produce both brief and lengthy responses as well as summaries from multiple sources.
With over 85% accuracy on USMLE-style questions and a pass rate of 72.3% on the MedMCQA dataset, which contains questions from India’s AIIMS and NEET medical exams, Med-PaLM 2 is the first language model to attain expert-level performance. The algorithm, which is tailored for medical queries, outperforms medical specialists while producing potentially dangerous answers just 5.9% of the time as opposed to 5.7% for human experts.