Kimi K3.1 Pro’s hacking capabilities have been deliberately nerfed. DeepSeek V… · @qwerty__cs · TwKit

@qwerty__cs

Kimi K3.1 Pro’s hacking capabilities have been deliberately nerfed. DeepSeek V4.1 Pro, Qwen 4.0 and GLM 5.5 are reportedly next. Cyber-related data was removed from both mid-training and reinforcement learning, while general coding tasks were retained. On Kimi’s internal exploit-development benchmark, the unfiltered checkpoint scored 90.5%; the final model scored 81.0% (pass@3).

曝光 190 · 评论 0 · 点赞 0 · 书签 0