How do you evaluate the quality of LLM outputs? What metrics and approaches are used for different types of tasks, and what are the limitations of automated evaluation?

Sample answer preview

Evaluating LLM outputs presents unique challenges because quality is multidimensional and often subjective. Different tasks require different evaluation approaches, and no single metric captures all aspects of response quality.