S surgehq.ai HellaSwag: 36% of this popular large language model benchmark contains errors 49 points echen 3 years ago 8 comments
