Output Length Effect on DeepSeek-R1's Safety in Forced Thinking

Li, Xuying, Li, Zhuo, Kosuga, Yuji · arXiv · 2025