Mor discusses how biases in data impact model performance, revealing a significant decrease in accuracy when models are tested on unseen annotators. The conversation delves into the role of advanced language models like Bert in capturing patterns and exploiting information content, shedding light on the importance of testing for biases in data collection.
