Data4Good 2025 - National Student Competition

AI is transforming education by providing adaptive tools, but it also risks spreading misinformation. Large Language Models (LLMs) often generate confident but incorrect answers, which can mislead students and harm trust in digital learning. This challenge asks participants to use the training set to explore, model, and classify AI-generated answers as factual, contradiction, or irrelevant. Then generate predictions for the held-out test set and document your full approach.

The dataset consists of question–answer pairs aimed at classifying whether the provided answer is factual, non-factual (contradiction), or irrelevant to the question. Each entry includes a Question (the prompt), Context (supporting information), Answer (the AI’s response), and Type, a categorical label with three possible values: factual (the answer is correct), contradiction (the answer is incorrect), or irrelevant (the answer is unrelated to the question). 

This competition uses a custom weighted confusion matrix to account for cost-based priorities. Each class receives an equal weighting to calculate your overall prediction performance. Thus, factual prediction score is 33.3%, contradiction classification 33.3%, and irrelevant (33.3%).  Your classification evaluation on the test set will be ranked among all teams in the competition. In addition to classification performance we are seeking a step-by-step approach in your notebooks.

Create a DataCamp account using your academic email, set up your DataLab workbook, and start collaborating with teammates. Publish and submit your work for review. You can submit it as many times as possible to check the scoring of your predictions. Prediction Guard will provide access to open-source LLMs via API for experimentation.