ChatGPT uninstalls surged by 295% after DoD deal
«Решетнев» рассказал о сборке спутника «Ямал-501»14:53
。PDF资料对此有专业解读
16:41, 4 марта 2026Мир
we assign a minterm id to each of these classes (e.g., 1 for letters, 0 for non-letters), and then compute derivatives based on these ids instead of characters. this is a huge win for performance and results in an absolutely enormous compression of memory, especially with large character classes like \w for word-characters in unicode, which would otherwise require tens of thousands of transitions alone (there’s a LOT of dotted umlauted squiggly characters in unicode). we show this in numbers as well, on the word counting \b\w{12,}\b benchmark, RE# is over 7x faster than the second-best engine thanks to minterm compressionremark here i’d like to correct, the second place already uses minterm compression, the rest are far behind. the reason we’re 7x faster than the second place is in the \b lookarounds :^).