The authors define 'experiments' for evaluating LLMs as being parametrised by five factors: (1) the number of data points processed, (2) the type of data ingested, (3) the model configuration (including model type, random seed, and temperature), (4) the prompt used, and (5) the number of clinicians required to review the data point for clinical errors.

Authors

• Person: Not available Organization: Nature
  A framework to assess clinical safety and hallucination rates of LLMs ...

Sources

• A framework to assess clinical safety and hallucination rates of LLMs ... www.nature.com Nature via serper

Referenced by nodes (1)

• Large Language Models concept