Post

Reviewing responses to an exam question that showed an intentionally small dataset of hand labeled sentiment alongside output of three models (two LLMs, one rule-based) The number of students questioning why you would ever hand-label a dataset is clearly something I need to address next year 😱