#laraoke #ai alignment failure #value misalignment #catastrophic misinterpretation #ai safety #unintended consequences #AI goal error