Medical LLM
A large language model trained or fine-tuned specifically for the healthcare domain, combining medical knowledge with clinical data. It is applied to tasks such as medical question answering, clinical reasoning, and electronic health record analysis.
Timeline (last 90 days)
Medical large language models are typically trained on a mixture of two types of data: instructional data (such as textbooks) and clinical data (such as patient records).
The optimal data mixing ratio is not fixed but varies with the downstream task's demands for knowledge versus clinical reasoning, and reasoning-intensive applications require a higher proportion of clinical data.
The study recommends that data curation for medical large language models should be application-driven rather than pursuing data scale or a standard formula.
The study recommends that evaluating medical models should not rely solely on knowledge-based benchmarks but should also incorporate clinical reasoning assessments.
All Facts (4)
Medical large language models are typically trained on a mixture of two types of data: instructional data (such as textbooks) and clinical data (such as patient records).
50%UnverifiedThe optimal data mixing ratio is not fixed but varies with the downstream task's demands for knowledge versus clinical reasoning, and reasoning-intensive applications require a higher proportion of clinical data.
50%UnverifiedThe study recommends that data curation for medical large language models should be application-driven rather than pursuing data scale or a standard formula.
50%UnverifiedThe study recommends that evaluating medical models should not rely solely on knowledge-based benchmarks but should also incorporate clinical reasoning assessments.
50%