Lilian Weng • 2/5/2024

Thinking about High-Quality Human Data

This technical article discusses the critical role of high-quality human-generated data in machine learning, particularly for tasks like classification and LLM alignment. It outlines best practices for data collection, including task design, annotator training, and quality assurance, and references historical and modern studies on crowdsourcing (e.g., Amazon Mechanical Turk) to illustrate the 'wisdom of the crowd' principle in data labeling.

0 comments

#Machine Learning #Data Quality #Rlhf