3 ms·LRM reasoning breaks down down past an unfamiliarity threshold, not "complexity"5 points by k1m 1y ago