LLM还是朴素贝叶斯？旧方法在新场景下的对比研究
原标题：LLMs or Naive Bayes? Old Gems or New Ways
AI 摘要
该研究对比了Complement Naive Bayes（CNB）与从27B到1T参数的多系列LLM在文本分类任务上的表现。结果显示LLM仅在无标注数据场景占优（Amazon Polarity上98.0%对88.2%），但存在数据污染风险；一旦有标注数据，CNB在AG News上达到89.1%，与零样本27B LLM持平并优于397B前沿模型，且CPU推理吞吐量比小LLM GPU批处理快40-486倍，每样本能耗低约两个数量级。作者认为对已有标注数据的资源受限HPC从业者，NB仍是最优选择，并提供了自动化模型选择的Kubernetes Helm operator。
正文节选
by LLMs or Naive Bayes? Old Gems or New Ways Abstract. Large language models (LLMs) prompt a recurring question in research computing: should classical methods like Naive Bayes (NB) be retired? We benchmark Complement Naive Bayes against zero-shot and few-shot LLMs spanning four model families and a range in scale (27B to a 1T-parameter mixture-of-experts) across text classification tasks. LLMs dominate only in zero-data regimes (98.0% vs 88.2% on Amazon Polarity sentiment), and even that win is