Comment on From GPT-4 to GPT-5: Measuring progress through MedHELM [pdf]Comments−xnx1yHave you looked at comparing to Google's foundation models or specialty medical models like MedGemma (https://developers.google.com/health-ai-developer-foundation...)?−fertrevinoOP1yThat would be an interesting extension. MedGemma isn't part of the original benchmark either [1]. Since Gemini 2.0 Flash is on 6th place, expectations are for MedGemma to achieve higher than that :)[1]https://crfm.stanford.edu/helm/medhelm/latest/#/leaderboard
Comments
Have you looked at comparing to Google's foundation models or specialty medical models like MedGemma (https://developers.google.com/health-ai-developer-foundation...)?
That would be an interesting extension. MedGemma isn't part of the original benchmark either [1]. Since Gemini 2.0 Flash is on 6th place, expectations are for MedGemma to achieve higher than that :)
[1]https://crfm.stanford.edu/helm/medhelm/latest/#/leaderboard