3 ms·AGI benchmark indicates whether a future AI model could cause catastrophic harm1 points by gmays 2y ago