by Osseus
Osseus builds medical and biological datasets, evaluations and reinforcement learning environments for frontier labs training AI models. It offers both off-the-shelf and custom datasets.
DatasetContact for pricing
Step 1: Decide whether you need an off-the-shelf dataset, a custom dataset or an interactive environment.
Step 2: Use the Get in touch link on the Osseus site to contact the team.
Step 3: Describe your model training or alignment work and the medical or biological areas involved.
Step 4: Agree the data or environment with Osseus for training and evaluation.
Step 5: Use the feedback and benchmarks to find remaining gaps in your model.
Medical and biological datasets covering patient histories and clinical conversations, pathology, molecular data and experimental results. They are available off the shelf or as custom builds.
They are multi-step medical and biological tasks in which a model gathers evidence, analyses data and uses clinical or scientific tools. The model gets feedback on accuracy, uncertainty and safe behaviour.
The site is aimed at frontier labs and others working on model training or alignment. They invite those teams to get in touch.
HealthBenchHealthBench is an OpenAI benchmark of 5,000 realistic health conversations, each graded with a physician-written rubric. It measures how well AI models respond in health settings.
WHO Global Health ObservatoryThe Global Health Observatory is the World Health Organization's data gateway for health statistics covering its 194 Member States. It groups more than 1000 health indicators by topic and country.