Naturalness
Project Naturalness focused on evaluating whether AI-generated responses aligned with a target behavioural style and response model. The work involved reviewing model outputs, identifying aligned and misaligned segments, and writing concise rationales explaining whether each highlighted section matched the expected standard. The evaluation required careful judgement of naturalness, instruction adherence, tone, usefulness, and response quality. Each assessment had to connect specific evidence from the response to the project rubric, while ensuring that the final score was consistent with the number and severity of aligned or misaligned observations.