Reasoning models can benefit from test-time scaling, but continued reasoning may become ineffective or harmful, a phenomenon termed overthinking. Evaluations…
机构:中科院
来源:arXiv 2610.13570 | AI4Papers 论文推荐平台