Recent work has identified a set of emergent symbolic mechanisms that support abstract reasoning in large language models, but it remains unclear what factor…
机构:普林斯顿
来源:arXiv 2611.13270 | AI4Papers 论文推荐平台