Large language models outperform humans at estimating society's everyday norms—but hybrids are even better