Revolutionising Interview Grading with Ethical AI

Study Details

• Data Sample: 10,000 candidates, 50,000 question-answer pairs from a pool of 3 million candidates, totalling over 1.3 billion words.

Analysis Process: An AI model was trained with specific interview data to create a tool that grades interviews and provides fairer, unbiased explanations (SAIGE).

Key Findings

• LLMs like GPT-4 and LLaMA-2, when used without fine-tuning, can carry over biases from their training data.
• InterviewLLM, fine-tuned with specific interview context data, performs better in generating relevant responses and grading.
• SAIGE, an extension of InterviewLLM, provides more accurate and fair grading and explanations compared to baseline models.

Key Takeaway
By customising LLMs with targeted interview data and using Behaviourally Anchored Rating Scales (BARS) for grading, companies can achieve more accurate and fair interview assessments. Sapia.ai’s approach shows how AI can enhance recruitment practices while ensuring ethical standards