LLM as judge - chunhualiao/public-docs GitHub Wiki

LLM-as-judge

• better to have multiple judges
• need calibration,
• with human raters,
• with cross-judge ensemble
• judge should not be an evaluated model (conflict of interest)