问HN:大型语言模型抛硬币——结果是正面
根据标题的内容。请让一个大型语言模型(LLM)抛硬币。正面在语言中是否存在偏见?<p>我了解大型语言模型的内部工作原理,这只是一个有趣的观察。<p>我们在生成标记时使用伪随机方法来选择更动态的语言,因此,从模型进行的一系列抛硬币理论上应该趋向于半随机性。
查看原文
As per the title. Ask an LLM to flip a coin. Is heads biased in language?<p>I'm aware how LLMs work internally, just an interesting observation.<p>We use pseudo-random to select more dynamic language during token generation, so a run of flips from the model should in theory tend towards semi-randomness.