Alex Banks: Google just announced Med-Gemini for medical tasks
Alex Banks, Co-Founder and CEO of
“Google just announced Med-Gemini for medical tasks.
It beats GPT-4 on all benchmarks:
Med-Gemini is a family of Gemini models fine-tuned for medical tasks.
The MedQA-USMLE benchmark is a dataset of USMLE-style multiple-choice questions used to evaluate the medical knowledge and reasoning capabilities of AI systems.
Med-Gemini set a new state-of-the-art of 91.1% on this benchmark.
It also outperformed GPT-4 models by an average margin of 44.5% on 7 multimodal benchmarks.
Something else that impressed me:
Med-Gemini results are preferred over human experts on long-form text-based tasks, including:
• Medical summarisation
• Doctor referral generation
• Medical simplification
AI outperforming doctors is nothing new.
We saw it in January when Google Research developed ‘AMIE’, a cutting-edge AI for medical reasoning and conversations.
AMIE outperformed both doctors alone and doctors equipped with AMIE for diagnostic accuracy.
My takeaway:
I see tremendous applications for Med-Gemini, not just for clinicians to administer better care but also for patients’ and medical students’ understanding of conditions.
Healthcare is proving to be one of the most exciting areas of AI development.
If you enjoy insights like this, follow me Alex Banks for more on AI.
P.S. Join 40,000+ others in my free newsletter to stay on the cutting edge of AI.”
Proceed to the video attached to the post.
Source: Alex Banks/LinkedIn