//

Home Innovation Google Google's Med-Gemini medical AI...

Google's Med-Gemini medical AI models surpass GPT-4


Google

Google's Med-Gemini medical AI models surpass GPT-4

Google unveiled a groundbreaking set of artificial intelligence (AI) models on Tuesday, specifically designed for the medical industry.

The tech company has released a pre-print version of its research paper explaining the abilities and methods of its AI models, called Med-Gemini. However, these models are not available for public use. According to the business, in benchmark testing, the AI models outperform the GPT-4 models. The AI model is good at understanding and studying research papers and health records because it can process long texts.

The study is available online at arXiv, a website for academic research. It is currently in the pre-print stage. Jeff Dean, Chief Scientist at Google DeepMind and Google Research, shared on X (formerly Twitter) that he is excited about how these models can help clinicians improve patient care and help patients understand their medical problems. He believes that one of the most significant application areas for AI will be healthcare.

The foundation of the Med-Gemini AI models is the Gemini 1.0 and 1.5 LLM. Med-Gemini-S 1.0, Med-Gemini-M 1.0, Med-Gemini-L 1.0, and Med-Gemini-M 1.5 are the four versions in total. All the models have multiple output formats, including text, image, and video. Additionally, the models are coupled with online search, which the business says has enhanced the models' ability to show results for complicated clinical reasoning tasks by self-training, making the models more factually accurate, dependable, and nuanced.

The company improved the AI model, so the chatbot can understand longer conversations and provide better answers, even with incorrect questions or long medical records. Google released data showing that Med-Gemini AI models outperformed OpenAI's GPT-4 models in text-based reasoning tasks on the GeneTuring dataset. Med-Gemini-L 1.0 performed 4.5% better than its previous model, achieving a 91.1% accuracy score on MedQA (USMLE). Beta testing and public access are currently unavailable. The corporation will make more improvements before releasing the model to the public.


Business News


Recommended News

Latest Magazine