A transformation lead at Fabrikam claims: “As long as we collect very large volumes of data, it doesn’t matter if it’s noisy or poorly labeled…

No, the statement is incorrect. This reflects a fundamental misunderstanding of how AI models are trained. The principle of "garbage in, garbage out" is highly relevant; models learn and amplify the patterns, biases, and errors present in their training data. While large models can tolerate some level of random noise, systemic issues like poor labeling, low-quality content, and inherent biases will degrade the model's performance, reliability, and safety. High-quality, clean, and relevant data is a critical prerequisite for building an effective and trustworthy generative AI model.

This is a "Yes/No" style question where 'B' represents "No, the statement is incorrect." The alternative would be "Yes," which is incorrect because it promotes the dangerous misconception that model scale can substitute for data quality, which is fundamentally false.

1. Ng, A. (2021). "Machine Learning Yearning". Stanford University. Chapter 8: "Establish a baseline, then iterate quickly". This and other chapters emphasize the critical importance of data quality over sheer quantity for model performance.

2. Microsoft Documentation, "What is data wrangling?". Azure Machine Learning documentation. Retrieved October 10, 2023. The entire premise of data preparation and wrangling tools is that data quality must be ensured before training.

3. Microsoft Learn, "Prepare data for AI". AI Transformation Leader Learning Path. Module 3, Unit 2: "Assess data readiness". This unit stresses that "The quality of your AI solution is directly dependent on the quality of your data."