Sampling several responses and voting over their answers can improve a language model's accuracy, but repeated answers limit the benefit of additional sample…
机构:ETH Zurich
来源:arXiv 2609.39560 | AI4Papers 论文推荐平台